WO2021109661A1 - 拥塞控制方法以及相关设备 - Google Patents

拥塞控制方法以及相关设备 Download PDF

Info

Publication number
WO2021109661A1
WO2021109661A1 PCT/CN2020/113775 CN2020113775W WO2021109661A1 WO 2021109661 A1 WO2021109661 A1 WO 2021109661A1 CN 2020113775 W CN2020113775 W CN 2020113775W WO 2021109661 A1 WO2021109661 A1 WO 2021109661A1
Authority
WO
WIPO (PCT)
Prior art keywords
congestion control
control parameter
network device
traffic
value
Prior art date
Application number
PCT/CN2020/113775
Other languages
English (en)
French (fr)
Inventor
晏思宇
郑晓龙
邓维山
夏寅贲
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to EP20895560.9A priority Critical patent/EP4054134A4/en
Publication of WO2021109661A1 publication Critical patent/WO2021109661A1/zh
Priority to US17/831,070 priority patent/US20220294736A1/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/12—Avoiding congestion; Recovering from congestion
    • H04L47/122—Avoiding congestion; Recovering from congestion by diverting traffic away from congested entities
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/28—Flow control; Congestion control in relation to timing considerations
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0852—Delays
    • H04L43/0864—Round trip delays
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0876—Network utilisation, e.g. volume of load or congestion level
    • H04L43/0882—Utilisation of link capacity
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/11—Identifying congestion
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/12—Avoiding congestion; Recovering from congestion
    • H04L47/127—Avoiding congestion; Recovering from congestion by using congestion prediction
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/25—Flow control; Congestion control with rate being modified by the source upon detecting a change of network conditions
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/10—Flow control; Congestion control
    • H04L47/29—Flow control; Congestion control using a combination of thresholds

Definitions

  • the embodiments of the present application relate to the field of network control, and in particular, to a congestion control method and related equipment.
  • TCP transmission control protocol
  • RDMA remote direct memory access
  • Congestion indicators mainly include delay, throughput and so on.
  • the congestion control parameters can be obtained through an artificial intelligence (AI) model installed on a network device, and the AI model can be obtained through a large number of historical traffic characteristics training using the initial AI model.
  • AI artificial intelligence
  • the process of using the AI model to obtain congestion control parameters is as follows: the network device collects the traffic characteristics on the network device, such as the egress forwarding rate, queue depth, etc., and the network device sends the collected traffic characteristics to the AI model for online reasoning.
  • the AI model is based on the traffic
  • the feature outputs congestion control parameters to the forwarding chip of the network device. Then, the forwarding chip uses the obtained congestion control parameters to control the flow of the network device.
  • AI models need to be trained from a large number of historical traffic characteristics.
  • the historical traffic characteristics are not extensive enough, the limited historical traffic characteristics cannot completely cover all traffic scenarios, which will lead to congestion control of the AI model output.
  • the parameters are not ideal, so the AI model has insufficient or even inadaptability, that is, the problem of scenario generalization.
  • the embodiments of the present application provide a congestion control method and related equipment, which can improve the scenario generalization of congestion control rules.
  • the first aspect of the embodiments of the present application provides a congestion control method.
  • a first congestion control rule is configured on the network device, and the network device can obtain the first congestion control parameter through the first congestion control rule. After the first device forwards traffic according to the first congestion control parameter, the network device can obtain the first traffic characteristic The first traffic characteristic includes statistical information generated when the first device forwards traffic according to the first congestion control parameter in the first period, and the network device may obtain the first reward value according to the acquired first traffic characteristic. After the network device obtains the first congestion control parameter through the first congestion control rule, the network device can use the first step length to modify the first congestion control parameter to obtain the second congestion control parameter, because the first step length is not equal to 0, so the first step length is not equal to 0. The second congestion control parameter is not equal to the first congestion control parameter.
  • the network device can obtain the second traffic characteristic, which includes the statistical information generated when the first device forwards the traffic according to the second congestion control parameter in the second period , The network device can obtain the second reward value according to the obtained second traffic characteristic.
  • the network device may determine whether the second reward value is greater than the first reward value, and if the network device determines that the second reward value is greater than the first reward value, the network device performs corresponding processing.
  • the network device may obtain the first traffic characteristic, and the first traffic characteristic includes that the first device performs the first congestion control parameter according to the first congestion control parameter in the first period.
  • the first congestion control parameter is obtained according to the first congestion control rule.
  • the network device can obtain the first return value according to the first traffic characteristic.
  • the network device can obtain the second congestion control parameter, and the network device can also obtain the second traffic characteristic.
  • the second traffic characteristic includes the statistical information generated when the first device forwards the traffic according to the second congestion control parameter in the second period.
  • the device can obtain a second reward value according to the second traffic characteristic.
  • the network device executes corresponding processing.
  • the network device uses the first step to modify the first congestion control parameter.
  • a second congestion control parameter that can obtain a larger reward value is obtained. Because the second reward value is greater than the first reward value, the second congestion control parameter is better than the first congestion control parameter. Therefore, the network device has a better response to the first congestion control rule.
  • the inference results have been optimized, and the generalization of the first congestion control rule has been improved.
  • the network device before the network device uses the first step to modify the first congestion control parameter to obtain the second congestion control parameter, the network device can obtain the first operating value and the second operating value of the first device, The first operating value is the operating value of the first device in the target period, and the second operating value is the operating value of the first device in the previous week of the target period.
  • the network device can also obtain the target threshold, which is set in advance The network device determines whether the absolute value of the difference between the first operating value and the second operating value is less than the target threshold. Only when the absolute value of the difference between the first operating value and the second operating value is less than the target threshold, the network device Only the step of using the first step to modify the first congestion control parameter to obtain the second congestion control parameter is performed.
  • the target period refers to the previous period of the first period.
  • the network device before the network device uses the first step to modify the first congestion control parameter and obtains the second congestion control parameter, the network device first determines whether it needs to use the first step to modify the first congestion control parameter, and the network device passes Determine whether the traffic forwarded by the first device tends to be stable to determine whether it is necessary to use the first step to modify the first congestion control parameter.
  • the method for the network device to determine whether the traffic forwarded by the first device tends to stabilize is: the network device obtains the first device For the first operating value and the second operating value in different periods, the network device determines whether the absolute value of the difference between the first operating value and the second operating value is less than the target threshold.
  • the network device determines that the traffic forwarded by the first device is stable, and the network device determines that it needs to use the first step to modify the first congestion control parameter.
  • the network device determines whether the traffic forwarded by the first device is Stable to determine whether it is necessary to use the first step to modify the first congestion control parameter, because when the traffic forwarded by the first device stabilizes, the subsequent modification of the first congestion control parameter by the network device will be more accurate, that is, the network device It can be determined that the result that the second reward value is greater than the first reward value is obtained by the network device using the first step to modify the first congestion control parameter.
  • the network device determines whether the first step of modifying the first congestion control parameter is required by determining whether the traffic forwarded by the first device is stable. Improve the accuracy of the program.
  • the network device can adjust the initial congestion control parameters only through the first congestion control rule, and adjusting the initial congestion control rule only through the first congestion control rule means that the network device does not use the step size to adjust the network device according to the first congestion control rule.
  • a congestion control parameter derived from a congestion control rule the network device directly uses the congestion control parameter derived from the first congestion control rule as a parameter for the first device to control traffic forwarding; the network device uses the first step to modify the first congestion control parameter
  • the network device can count the number of consecutive adjustments that the network device only adjusts the initial congestion control parameter through the first congestion control rule. When the number of consecutive adjustments is greater than the set threshold N, the network device will use the first congestion control parameter. Modify the first congestion control parameter in one step. Accordingly, the first congestion control parameter is obtained by continuously adjusting the initial congestion control parameter N+1 times according to the first congestion control rule.
  • the network device before the network device uses the first step to modify the first congestion control parameter and obtains the second congestion control parameter, the network device can count the continuous adjustment of the network device only through the first congestion control rule to adjust the initial congestion control parameter When the number of consecutive adjustments is greater than the set threshold N, the network device can use the first step to modify the first congestion control parameter.
  • the network device After the network device obtains the congestion control parameter according to the first congestion control rule, the network device It is not necessary to use the step size to modify the congestion control parameters each time, so the network device may not use the step size to adjust the congestion control parameters obtained by the network device according to the first congestion control rule, but directly use the congestion control parameter obtained by the first congestion control rule.
  • the control parameters are used as the parameters for the first device to control traffic forwarding, but the judgment mechanism of the network device may not be perfect.
  • the judgment mechanism of the network device refers to: the network device chooses to use the congestion control parameter after the step size modification as the first device to control the flow forwarding
  • the parameter is still to choose to directly use the congestion control parameter derived from the first congestion control rule as the selection condition of the parameter for the first device to control traffic forwarding; if the judgment mechanism of the network device is not perfect, it may cause the network device to continue to choose to directly congest the first congestion
  • the congestion control parameters derived from the control rules are used as the parameters for the first device to control traffic forwarding, and the congestion control parameters modified by the step size are not selected as the parameters for the first device to control traffic forwarding, that is, the network device does not use the first step length
  • the step of modifying the first congestion control parameter to obtain the second congestion control parameter because when the network device determines that the number of consecutive adjustments is greater than the set threshold N, the network device can use the first
  • Q is the congestion control parameter
  • A is the rate
  • B is the set delay.
  • F(A, B) is a function related to A and B.
  • the first congestion control rule is specifically defined as the first formula, and some parameters in the formula are specifically defined, where A is the rate, B is the set delay, and B is the pre-set value.
  • A is the rate acquired by the network device.
  • the network device can use the acquired rate and the set delay to obtain the congestion control parameters through the first formula. When the acquired rate changes, the congestion control parameters may also change accordingly.
  • the network device obtains different congestion control parameters according to different rates, which realizes the dynamic control of the forwarding traffic by the first device, thereby improving the feasibility of the solution.
  • the relationship between A and B is specifically defined, which improves the feasibility of the solution.
  • the network device uses the first step to modify the first formula to obtain the second formula.
  • the network device uses the first step to modify the first congestion control parameter and obtains the second congestion control parameter
  • the network device uses the first step to modify the first congestion control parameter.
  • the network device uses the first step to modify the first congestion control parameter. In the first formula, the network device not only optimizes the first congestion control parameter derived from the first congestion control rule, but also optimizes the first congestion control rule, so the generalization of the first congestion control rule is improved from the root cause.
  • the network device uses the second formula to obtain the third congestion control parameter according to the acquired second traffic characteristic, and the third congestion control parameter is used by the first device to control the forwarded traffic.
  • the network device can use the second formula to obtain the third congestion control parameter according to the second traffic characteristics, and the third congestion control parameter can be used to control the first congestion control parameter.
  • the congestion index of the first device is improved, and the generalization of the first formula is improved.
  • the first step length is a percentage value.
  • the first step length is a percentage value, because in different traffic models, the value of the first congestion control parameter may vary greatly. If the first step length is a specific value, such as 20, when the first When the congestion control parameter is 1000, the first step to modify the first congestion control parameter will appear too small; when the first congestion control parameter is in the range of 0-30, the first congestion control parameter is 10, and the first step is At 20:00, the modification of the first congestion control parameter in the first step will be too large. Therefore, the first step can use a percentage value to remove the interference caused by the change in the size of the first congestion control parameter itself. Therefore, in practical applications, The first step can be applied to different first congestion control parameters, so the adaptability of the solution in different scenarios is improved, that is, the generalization of the solution is improved.
  • the network device uses the second step size to modify the second congestion control parameter to obtain the third congestion control parameter.
  • the network device can obtain the third traffic characteristic, which includes the statistical information generated when the first device forwards the traffic according to the third congestion control parameter in the third period, and the network device can obtain the statistics according to the To obtain the third flow characteristic, the third return value is obtained.
  • the network device uses the second step to modify the first congestion control parameter forward.
  • the second congestion control parameter is to obtain the third congestion control parameter.
  • the second congestion control parameter is better than the first congestion control parameter.
  • the second congestion control parameter is The control parameters may not be the optimal congestion control parameters.
  • the network device makes further optimization attempts on the basis of the second congestion control parameter, thus increasing the probability of the network device obtaining better congestion control parameters and increasing the first congestion control parameter.
  • the degree of generalization of the control rules is referred to modify the first congestion control parameter forward.
  • the network device can use the third step to reversely modify the second congestion control parameter to obtain the fourth congestion control parameter.
  • the network device can obtain the fourth traffic characteristic, and the fourth traffic characteristic includes the statistical information generated when the first device forwards the traffic according to the fourth congestion control parameter in the fourth cycle, and the network device can obtain the statistics according to the The fourth flow characteristic of the, obtain the fourth return value.
  • the second congestion control parameter is a value.
  • the network device can modify this value to make the second congestion control parameter larger or smaller.
  • the network device attempts to increase or decrease the second congestion control parameter to obtain After the third congestion control parameter, if the third return value is less than the second return value, the network device can determine that the second congestion control parameter is better than the third congestion control parameter, and the network device can determine that the previous attempt to change the second congestion control parameter The behavior of getting bigger or smaller is wrong. Therefore, the network device uses the third step to reversely modify the second congestion control parameter to obtain the fourth congestion control parameter.
  • the reverse modification defined here is only for the purpose of the aforementioned forward modification. Distinguish, unrestricted reverse modification means that the network device performs a reduction operation on the second congestion control parameter.
  • the network device attempts to obtain a third congestion control parameter that is better than the second congestion control parameter and fails.
  • an attempt is made to obtain a fourth congestion control parameter that is better than the second congestion control parameter, thereby increasing the probability that the network device obtains a better congestion control parameter, and increasing the generalization degree of the first congestion control rule.
  • the network device uses the third step to modify the second congestion control parameter to obtain the fourth congestion control parameter. After the first device forwards the traffic according to the fourth congestion control parameter, the network device can obtain the fourth traffic.
  • the fourth traffic characteristic includes statistical information generated when the first device forwards traffic according to the fourth congestion control parameter in the fourth cycle.
  • the network device can obtain the fourth return value according to the acquired fourth traffic characteristic. If the fourth reward value is greater than the second reward value, the network device uses the third step length and the first step length to modify the first congestion control rule to obtain the second congestion control rule.
  • the network device uses the third step size to modify the second congestion control parameter and obtains the fourth congestion control parameter, determines that the fourth return value is greater than the second return value, that is, the fourth congestion control parameter is better than The second congestion control parameter, and the second reward value is greater than the first reward value, that is, the second congestion control parameter is better than the first congestion control parameter.
  • the network device uses the third step length and the first step length to modify the first congestion control Rules, in which, after the network device determines that the fourth congestion control parameter is better than the second congestion control parameter, the network device uses the third step length and the first step length to modify the first congestion control rule, and the network device not only optimizes twice according to the first
  • the first congestion control parameter derived from a congestion control rule also optimizes the first congestion control rule, so the generalization of the first congestion control rule is improved from the root cause.
  • the network device obtains a fourth congestion control parameter that is superior to the second congestion control parameter, and uses the third step length and the first step length to modify the first congestion control rule to obtain the second congestion control parameter. After the rule, the network device uses the second congestion control rule to generate a new congestion control parameter of the first device according to the fourth traffic characteristic, and the new congestion control parameter is used by the first device to control the forwarded traffic.
  • the network device can use the second congestion control rule to obtain the new congestion control parameter according to the fourth traffic characteristic.
  • the control parameter can be used to control the traffic forwarded by the first device, where the network device uses the new congestion control parameter to obtain the new congestion control parameter, that is, the network device uses the second congestion control rule to replace the first congestion control rule, and the first congestion control rule
  • the second congestion control rule is optimized from the first congestion control rule, so the congestion index of the first device is improved, and therefore the generalization of the first congestion control rule is improved.
  • the second step length is greater than the first step length.
  • the second step length is greater than the first step length.
  • the network device uses the first step length to modify the first congestion control parameter, the network device obtains the second congestion control parameter that is better than the first congestion control parameter. This proves the necessity and correctness of the network device to modify the first congestion control parameter. Therefore, the network device can use the second step size greater than the first step size when modifying the second congestion control parameter by the step size, where, Because the second step length is greater than the first step length, within the value range of the congestion control parameter, the network device can determine the optimal congestion control parameter considered by the network device more quickly, thus reducing the convergence time.
  • a third congestion control rule is configured on the network device, and the network device can obtain the fifth congestion control parameter through the third congestion control rule.
  • the network device may obtain the fifth flow characteristic, and the fifth flow characteristic includes the statistical information generated when the second device forwards the flow according to the fifth congestion control parameter in the first period, and the network device may obtain the fifth flow characteristic according to the obtained fifth flow characteristic.
  • the fifth return value after the network device obtains the fifth congestion control parameter through the third congestion control rule, the network device can use the fourth step to modify the fifth congestion control parameter to obtain the sixth congestion control parameter, because the first step is not Equal to 0, so the sixth congestion control parameter is not equal to the fifth congestion control parameter.
  • the network device can obtain the sixth traffic feature, which includes the second device’s In the second cycle, according to the statistical information generated when the traffic is forwarded according to the sixth congestion control parameter, the network device can obtain the sixth return value according to the acquired sixth traffic characteristics; the network device can determine the sum of the sixth return value and the second return Whether it is greater than the sum of the fifth reward value and the first reward value, if the network device determines that the sum of the sixth reward value and the second reward value is greater than the sum of the fifth reward value and the first reward value, the network device performs the corresponding corresponding Processing.
  • the network device may obtain the fifth flow characteristic, and the fifth flow characteristic includes the second device according to the fifth congestion control parameter in the first period. Statistics generated when forwarding traffic.
  • the fifth congestion control parameter is obtained according to the third congestion control rule.
  • the network device can obtain the fifth return value according to the fifth traffic characteristic.
  • the network device can obtain the sixth congestion control parameter, and the network device can also obtain the sixth traffic characteristic.
  • the sixth traffic characteristic includes the statistical information generated when the second device forwards the traffic according to the sixth congestion control parameter in the second period.
  • the device can obtain the sixth reward value according to the sixth traffic characteristic; after the network device obtains the first reward value, the second reward value, the fifth reward value, and the sixth reward value, the network device can determine the sixth reward value and Whether the sum of the second reward is greater than the sum of the fifth reward and the first reward, if the network device determines that the sum of the sixth reward and the second reward is greater than the sum of the fifth reward and the first reward, the network device executes According to the corresponding processing, in the flow control of the network, the flow between the first device and the second device may affect each other. Therefore, the network device comprehensively considers the sum of the return values on the two devices, and the network device uses the second device. Modify the first congestion control parameter in one step to obtain the second congestion control parameter.
  • the network device uses the fourth step to modify the fifth congestion control parameter to obtain the sixth congestion control parameter, because the sixth return value is different from the second return value.
  • the sum is greater than the sum of the fifth return value and the first return value. Therefore, the profit of the network device modifying the first congestion control parameter and the second congestion control parameter this time is positive. Therefore, when the flow control of multiple devices is forwarded, the first The scenario generalization of the first congestion control rule and the second congestion control rule.
  • the network device uses the fourth step to modify the third congestion control rule, the network device uses the first step to modify the first congestion control rule.
  • the network device uses the fourth step to modify the fifth congestion control parameter and obtains the fifth congestion control parameter, determines that the sum of the sixth reward value and the second reward is greater than the fifth reward value and the first reward value.
  • the sum of the return value that is, the comprehensive return value generated by the sixth congestion control parameter and the second congestion control parameter is better than the comprehensive return value of the fifth congestion control parameter and the first congestion control parameter.
  • the network device uses the first step to modify The first congestion control rule, the network device uses the fourth step size to modify the third congestion control rule, wherein the network device is determining that the sum of the sixth reward value and the second reward is greater than the sum of the fifth reward value and the first reward value After that, the network device modifies the first congestion control rule and the third congestion control rule.
  • the network device not only optimizes the first congestion control parameter derived from the first congestion control rule and the fifth congestion control parameter derived from the third congestion control parameter. , It also optimizes the first congestion control rule and the third congestion control rule, so the generalization of the congestion control rule is improved from the root cause.
  • the network device uses the second step size to modify the second congestion control parameter, Obtain the third congestion control parameter, the network device uses the fifth step to modify the sixth congestion control parameter to obtain the seventh congestion control parameter, the first device forwards traffic according to the third congestion control parameter, and the second device according to the seventh congestion control parameter
  • the network device can obtain the third traffic feature and the seventh traffic feature.
  • the third traffic feature includes the statistical information generated when the first device forwards the traffic according to the third congestion control parameter in the third period.
  • the seventh traffic The characteristics include statistical information generated when the second device forwards traffic according to the third congestion control parameter in the third period, the network device obtains a third reward value according to the third traffic characteristic, and obtains a seventh reward value according to the seventh traffic characteristic.
  • the network device uses the second step size to modify the second congestion control parameter to obtain the third Congestion control parameters.
  • the network device uses the fifth step to modify the sixth congestion control parameter to obtain the seventh congestion control parameter.
  • the combined benefit of the sixth congestion control parameter and the second congestion control parameter is greater than that of the fifth congestion control parameter and the first congestion control parameter.
  • the comprehensive benefits of congestion control parameters that is, the return value generated by the combination of the sixth congestion control parameter and the second congestion control parameter is greater than the return value generated by the combination of the fifth congestion control parameter and the first congestion control parameter, and the network equipment is getting better After the combination of the first congestion control parameter and the fifth congestion control parameter, but because the combination of the second congestion control parameter and the sixth congestion control parameter may not be the optimal congestion control parameter combination, the network device is in the second congestion control Further optimization attempts are made on the basis of the combination of the parameters and the sixth congestion control parameter, thus increasing the probability that the network device obtains better than the combination of the second congestion control parameter and the sixth congestion control parameter, and improving the first congestion control rule and the third congestion control parameter.
  • the degree of generalization of congestion control rules is made on the basis of the combination of the parameters and the sixth congestion control parameter, thus increasing the probability that the network device obtains better than the combination of the second congestion control parameter and the sixth congestion control parameter, and improving the first congestion control rule and the third congestion control parameter.
  • the second aspect of the embodiments of the present application provides a congestion control device, which includes multiple functional modules.
  • the multiple functional modules may all be software modules or hardware modules, or a combination of software modules and hardware modules. Modules can be divided differently according to the implementation, and the methods in the above-mentioned first aspect and various implementation modes can be used as the criterion.
  • a third aspect of the embodiments of the present application provides a congestion control device, including a memory and a processor.
  • the memory is used to store a program
  • the processor is configured to execute the program in the memory, so that the congestion control device executes the method described in the first aspect and any one of the implementation manners thereof.
  • the fourth aspect of the embodiments of the present application provides a computer storage medium.
  • the computer storage medium stores instructions.
  • the instructions When the instructions are executed on a computer, the computer executes the first aspect and any one of the implementations described above. The method described in the way.
  • the fifth aspect of the embodiments of the present application provides a computer program product.
  • the computer program product When the computer program product is executed on a computer, the computer executes the method described in the first aspect and any one of its implementation manners.
  • Figure 1 is a schematic diagram of a network framework of an embodiment of the application
  • FIG. 2 is a schematic flowchart of a congestion control method in an embodiment of the application
  • FIG. 3 is a schematic diagram of another flow of a congestion control method in an embodiment of this application.
  • FIG. 4 is another schematic flow chart of the congestion control method in an embodiment of this application.
  • FIG. 5 is a schematic diagram of another flow of the congestion control method in an embodiment of this application.
  • FIG. 6 is a schematic structural diagram of a congestion control device in an embodiment of the application.
  • FIG. 7 is another schematic structural diagram of the congestion control device in an embodiment of the application.
  • Fig. 8 is a schematic structural diagram of a congestion control device in an embodiment of the application.
  • the embodiments of the present application provide a congestion control method and related equipment, which are used in the field of network control, and can improve the generalization of congestion control rules.
  • AI model is an algorithm, which refers to a model that can be solved through input data after training with historical data.
  • the AI model can be either a mainstream deep neural network model or a traditional machine learning model.
  • the network node In the network, when there is traffic passing through a certain network node, the network node will generate flow. In order to maintain the stability of the flow, that is, in order to meet certain conditions, the flow of the network node is controlled.
  • the network node can be a network device, such as a server, a switch, a router, and so on.
  • network technologies such as TCP and RDMA are widely used in wide area networks, data centers and other fields, and these network technologies have higher and higher demands on the network, among which throughput and delay are the main congestion indicators.
  • TCP network devices drop packets according to a drop probability parameter to perform flow control
  • RDMA RDMA network
  • an explicit congestion notification (ECN) waterline is dynamically adjusted to meet throughput and delay.
  • ECN explicit congestion notification
  • the ECN mechanism is widely used in high-performance data center networks based on enhanced RoCE (RDMA over Converged Ethernet), and a reasonable ECN configuration plays a key role in network flow control.
  • the switch has three ECN values that can be configured for the queue, which are the upper line, the lower line, and the maximum mark probability.
  • the two thresholds of the queue length define the marking probability, and the two thresholds are the lower line and the upper line respectively.
  • ECN When the queue length is lower than the threshold waterline, ECN will not be marked, which is equivalent to the actual marking probability of 0; when the queue length exceeds the threshold waterline, all network packets transmitted from the queue will be ECN Marking is equivalent to the actual marking probability of 1.
  • the data packet When the queue length is between two thresholds, the data packet will be ECN marked with a probability of linear growth with the queue length.
  • the switch Take the switch as an example to illustrate the meaning of the ECN mark.
  • the switch is configured with ECN waterline and downline. When the switch port is congested, it will determine whether to mark the packet with ECN according to the ECN threshold.
  • the message generates a congestion notification packet (CNP) message to inform the source end, and the source end network card reduces the occurrence rate according to the number of CNP messages, thereby avoiding congestion.
  • CNP congestion notification packet
  • the diversity of business and network traffic will produce different network traffic models. Under different network traffic models, how network equipment dynamically adjusts and accurately adapts congestion control parameters according to changes in traffic so as to ensure network performance has become an important challenge in the network.
  • the network framework of the embodiment of the present application includes: a network device 101, a first device 102, and a second device 103; wherein, the network device 101 is connected to the first device 102 and the second device 103, respectively.
  • the manner in which the network device 101 is connected to the first device 102 and the manner in which the network device 101 is connected to the second device 103 may be connected through a wired network or through a wireless network.
  • the network device 101 can be connected to more devices.
  • the main function of the network device 101 is to obtain the traffic characteristics of the first device 102 or the first device 102 and the second device 103.
  • the traffic characteristics include the first device 102, or the first device 102 and the second device 103 to forward traffic according to congestion control parameters.
  • the network device 101 uses congestion control rules to generate congestion control parameters according to the acquired traffic characteristics; the network device 101 can also calculate the return value according to the acquired traffic characteristics to evaluate the first device 102 or the first device 102 or the first device 102.
  • the congestion status of the device 102 and the second device 103 is to obtain the traffic characteristics of the first device 102 or the first device 102 and the second device 103.
  • the main function of the first device 102 is to use the acquired congestion control parameters to control the forwarding of traffic, and to generate traffic characteristics according to the forwarding of the traffic.
  • the function of the second device 103 is similar to that of the first device 102.
  • the embodiment of the present application may not have the network device 101.
  • the first device 102 can complete all the functions of the network device 101.
  • the embodiment of the present application may have no network device 101 and no second device 103, the first device 102 does not need to jointly control the forwarding of traffic with the second device 103, and the first device 102 completes all the functions of the network device 101.
  • FIG. 2 is a schematic flowchart of a congestion control method in an embodiment of this application.
  • step 201 the network device obtains a first flow characteristic, and obtains a first reward value according to the first flow characteristic data.
  • the network device is configured with a first congestion control rule.
  • the network device can obtain the first congestion control parameter according to the first congestion control rule, and the network device can obtain the first congestion control parameter according to the first congestion control rule.
  • the network device can obtain the first traffic characteristic.
  • the first traffic characteristic includes the statistical information generated when the first device forwards traffic according to the first congestion control parameter in the first period.
  • the network device can obtain the first traffic characteristic according to the first congestion control parameter. Features, get the first return value.
  • step 202 the network device uses the first step to modify the first congestion control parameter to obtain the second congestion control parameter.
  • the network device uses the first step to modify the first congestion control parameter to obtain the second congestion control parameter, and the second congestion control parameter is not equal to the first congestion control parameter.
  • step 203 the network device obtains a second flow characteristic, and obtains a second reward value according to the second flow characteristic.
  • the network device uses the first step to modify the first congestion control parameter. After obtaining the second congestion control parameter, the network device can obtain the second flow characteristic.
  • the second flow characteristic includes the first device according to the second congestion control parameter in the second period. Based on the statistical information generated when the traffic is forwarded, the network device can obtain the second return value according to the second traffic characteristic.
  • step 204 if the second reward value is greater than the first reward value, the network device performs corresponding processing.
  • the network device can determine whether the second reward value is greater than the first reward value. If the network device determines that the second reward value is greater than the first reward value, the network device executes the corresponding Processing.
  • the network device can perform a variety of different processing.
  • the processing under such conditions is collectively referred to as corresponding processing in this application.
  • the corresponding processing performed by the network device may be regarded as a corresponding operation of accepting the modification of the first congestion control parameter.
  • the network device may obtain the first flow characteristic, and the first flow characteristic includes the first device according to the first congestion control parameter in the first period.
  • the first congestion control parameter is obtained according to the first congestion control rule.
  • the network device can obtain the first return value according to the first traffic characteristic.
  • the network device can obtain the second congestion control parameter, and the network device can also obtain the second traffic characteristic.
  • the second traffic characteristic includes the statistical information generated when the first device forwards the traffic according to the second congestion control parameter in the second period.
  • the device can obtain a second reward value according to the second traffic characteristic.
  • the network device executes corresponding processing.
  • the network device uses the first step to modify the first congestion control parameter.
  • a second congestion control parameter that can obtain a larger reward value is obtained. Because the second reward value is greater than the first reward value, the second congestion control parameter is better than the first congestion control parameter. Therefore, the network device has a better response to the first congestion control rule. The inference result is optimized, thus improving the generalization of the first congestion control rule.
  • the first congestion control parameter may be one parameter, and the first congestion control parameter may also be multiple parameters, which are described separately below.
  • the first congestion control parameter is a parameter.
  • FIG. 3 is a schematic flowchart of another embodiment of the congestion control method provided by this application.
  • step 301 the network device obtains the first congestion control parameter.
  • the network device is configured with the first congestion control rule.
  • the first congestion control rule is the first formula.
  • the network device can input the collected flow characteristics into the first formula to obtain the first formula.
  • the first congestion control parameter output by a formula in order to facilitate the understanding of the congestion control method in this embodiment, this embodiment uses the following formula as the first formula:
  • Q is the congestion control parameter
  • A is the rate
  • B is the set delay
  • Rate A can be the forwarding rate of the first device or a certain queue of the first device.
  • the forwarding rate is a type of traffic characteristic.
  • the traffic characteristic refers to the statistical information generated when the first device forwards traffic, such as the number of outgoing message bytes. , Number of outgoing packets, queue depth, number of packets marked by ECN, throughput information, number of lost packets, etc.
  • the set delay B is a value set in advance.
  • the congestion control configuration personnel of the first device can set the value of the set delay based on experience or a template.
  • the value of the set delay reflects the configuration personnel’s Delay requirements in the congestion indicator of the device.
  • the specific process of obtaining the first congestion control parameter can be as follows: the network device collects the rate A of the first device in the previous cycle, obtains the preset set delay B, and then sets the rate A and the set delay B into the first Formula to obtain the first congestion control parameter Q.
  • the network device can use the pre-set rate, or not acquire the first congestion control parameter in this cycle, wait for the next cycle to acquire the rate in this cycle, and use the rate in this cycle.
  • the rate obtains the first congestion control parameter to control the flow of the first device in the next cycle.
  • the first congestion control parameter may be the marking probability and the discarding probability.
  • the first congestion control rule may be an AI model.
  • an AI model may be selected as the first congestion control rule.
  • this embodiment takes the first congestion control rule as the first formula as an example for description.
  • the network device may be the first device, or may be a device other than the first device.
  • the network device obtains the first traffic characteristic, the first traffic characteristic includes the statistical information generated when the first device forwards the traffic according to the first congestion control parameter in the first period, and the network device obtains the first traffic characteristic according to the first traffic characteristic. A return value.
  • the network device can obtain the first flow characteristic.
  • the first flow characteristic includes the data generated when the first device forwards the flow according to the first congestion control parameter in the first period.
  • the first traffic characteristic can specifically be the statistical information of the first device, or the statistical information of the port of the first device, or the statistical information of the queue of the first device; the network device can obtain the first return according to the first traffic characteristic value.
  • the first device uses the first congestion control parameter to control the traffic forwarded by the first device in the first period, and the first device uses the second congestion control parameter to control the traffic in the first period.
  • the traffic forwarded by the first device is controlled in the second cycle.
  • the first traffic feature includes the statistical information generated when the first device uses the first congestion control parameter to forward traffic in the first cycle.
  • the second traffic feature includes the first device in the first cycle. The statistical information generated when the first congestion control parameter is used to forward traffic within a week, and so on.
  • the period duration of the first period and the second period may be the same or different.
  • the first period and the second period may have an interval, may be adjacent, or may overlap.
  • the network device can obtain the first traffic characteristic, which may be a value at a certain moment in the first cycle , It may also be a processed value.
  • the first flow characteristic may be an average value in the first period.
  • the network device when the network device is the first device, the network device directly issues the first congestion control parameter to the forwarding chip, so that the forwarding chip uses the first congestion control parameter to control the traffic forwarded by the first device in the first cycle.
  • the network device when the network device is a device other than the first device, after the network device obtains the first congestion control parameter, the network device sends the first congestion control parameter to the first device, so that the first device is in the first cycle
  • the first congestion control parameter is used to control the traffic forwarded by the first device in the first cycle.
  • Congestion indicators will include multiple indicators, such as latency and throughput. In order to facilitate the evaluation of the pros and cons of the first device's different cycles of congestion indicators, the multiple indicators are integrated into one indicator. This indicator is Return value.
  • the first flow characteristic undergoes preprocessing to obtain throughput or delay.
  • the queue depth may be converted into delay through preprocessing, for example, when the first flow characteristic is the outgoing text
  • the number of knots can be preprocessed to convert the number of bytes of the outgoing message into a throughput, and then the first return value can be calculated according to the throughput or delay obtained after the preprocessing.
  • the network device can use the following algorithm to obtain the first return value:
  • R is the first return value
  • J is throughput
  • K is delay
  • m and n are weighting coefficients.
  • the network device can use the following algorithm to obtain the first return value:
  • R is the first return value
  • J is the throughput
  • K is the delay
  • m and n are the weighting coefficients
  • L is the service performance index
  • v is the weighting coefficient.
  • the network device obtains the first operating value and the second operating value of the first device, and determines whether the difference between the first operating value and the second operating value is less than a target threshold.
  • the target threshold is a value set in advance.
  • the network device can use the first step to modify the first congestion control parameter.
  • the first operating value is the forwarding rate of the first device in the target period
  • the target period refers to the previous period of the first period
  • the second operating value is the forwarding rate of the first device in one week before the target period.
  • the first operating value is the forwarding rate of the first queue of the first device in the target period
  • the second operating value is the forwarding rate of the first queue of the first device in a week before the target period.
  • the first running value is the queue depth of the first queue of the first device in the target period
  • the target period refers to the previous period of the first period
  • the first running value is the first queue of the first device in the target period.
  • the target threshold may also be different.
  • the target threshold is the first threshold
  • the first operating value is that of the first device
  • the target threshold is the second threshold
  • the first threshold is not equal to the second threshold.
  • the network device may adjust the initial congestion control parameter only through the first congestion control rule, and adjusting the initial congestion control rule only through the first congestion control rule means that the network device does not use the step size to adjust the network device according to the first congestion control rule.
  • the network device directly uses the congestion control parameter obtained by the first congestion control rule as the parameter for the first device to control traffic forwarding; the network device uses the first step to modify the first congestion control parameter to obtain the second congestion control parameter.
  • the network device may count the number of consecutive adjustments of the initial congestion control parameter by the network device only through the first congestion control rule.
  • step 304 if the difference between the first operating value and the second operating value is less than the target threshold, the network device uses the first step to forwardly modify the first congestion control parameter to obtain the second congestion control parameter.
  • the network device uses the first step to forwardly modify the first congestion control parameter to obtain the second congestion control parameter.
  • the forward modification is only to distinguish it from the reverse modification described later. It does not mean that the first congestion control parameter can only be added forward.
  • the first congestion control parameter is 100
  • the first step length is 20, and the network device
  • the second congestion control parameter can be 80 or 120.
  • the second congestion control parameter is 120 is taken as an example for description.
  • the first step length is a percentage value, because in the network, the first congestion control parameter may vary greatly. If the first step length is a specific value, such as 20, when the first congestion control parameter is 1000, The first step is too small to modify the first congestion control parameter. Therefore, the first step can use a percentage value to remove the interference caused by the change of the first congestion control parameter. When the first congestion control parameter is 100, the first step When the length is 20%, the network device uses the first step to forwardly modify the first congestion control parameter, and the obtained second step can be 80 or 120.
  • the target threshold can be adjusted according to the size of the network fluctuation of the first device.
  • the network device reduces the target threshold.
  • the network fluctuates greatly, that is, the traffic of the first device changes greatly, and the network device increases the target threshold.
  • the network device may not obtain the first operating value and the second operating value of the first device, and the network device It can be determined whether the number of consecutive adjustments is greater than the set threshold N.
  • the network device executes the first long forward modification of the first congestion control parameter to obtain the second congestion. Steps to control parameters.
  • the network device executes the step of using the first step to forwardly modify the first congestion control parameter to obtain the second congestion control parameter, which includes the following situations:
  • the network device executes the first step of forward modification of the first congestion control parameter , The step of obtaining the second congestion control parameter.
  • the network device executes the first step of forward modification of the first congestion control parameter , The step of obtaining the second congestion control parameter.
  • the network device executes the first step of forward modification of the first congestion control parameter , The step of obtaining the second congestion control parameter.
  • step 303 may not be performed.
  • the network device may not need to confirm that the difference between the first operating value and the second operating value is less than the target threshold, and directly Use the first step to forwardly modify the first congestion control parameter to obtain the second congestion control parameter.
  • the network device obtains a second flow characteristic, and the second flow characteristic includes the statistical information generated when the first device forwards the flow according to the second congestion control parameter in the second period; the network device according to the second flow characteristic, Obtain the second return value.
  • the network device uses the first step to positively modify the first congestion control parameter.
  • the network device can obtain the second flow characteristic.
  • the second flow characteristic includes the first device according to the first congestion control parameter in the second period.
  • the second congestion control parameter is the statistical information generated when the traffic is forwarded.
  • the second traffic characteristic may specifically be the statistical information of the first device, or the statistical information of the port of the first device, or the statistical information of the queue of the first device; the network device obtains the first device’s statistical information.
  • the network device can obtain a second reward value according to the acquired second traffic characteristic, and the algorithm of the second reward value is similar to the algorithm of the first reward value in step 302.
  • the second flow characteristic may be the value of the first device at a certain moment in the second period, or a processed value.
  • the second flow characteristic may be the average value of the first device in the second period.
  • the network device when the network device is the first device, the network device directly issues the second congestion control parameter to the forwarding chip, so that the forwarding chip controls the traffic forwarded by the first device in the second cycle.
  • the network device when the network device is a device other than the first device, after the network device obtains the second congestion control parameter, the network device sends the second congestion control parameter to the first device for the first device to use The second congestion control parameter controls the traffic forwarded by the first device in the second cycle.
  • step 306 the network device confirms whether the second reward value is greater than the first reward value.
  • the network device After the network device obtains the second reward value and the first reward value, the network device confirms whether the second reward value is greater than the first reward value.
  • step 307 if the second reward value is greater than the first reward value, the network device uses the first step to modify the first formula to obtain the second formula, and uses the second formula to obtain the congestion control parameter.
  • the network device uses the first step to modify the first formula to obtain the second formula, and uses the second formula to obtain the congestion control parameter.
  • the second formula may be the following formula:
  • Q is the congestion control parameter
  • A is the rate
  • B is the set delay
  • C1 is the first step of the percentage value.
  • the network device uses the rate of the first device in the first cycle to set the delay and the first step length, and the congestion control parameter can be obtained through the second formula.
  • the congestion control parameter is used as the first congestion control parameter
  • the second formula is used as the congestion control parameter. For the first formula, return to step 302.
  • the network device determines that the second reward value is greater than the first reward value, the network device does not need to modify the first formula in the first step to obtain the second formula, but uses the second step to continue forward modification
  • the second congestion control parameter obtains the third congestion control parameter, and the network device obtains the third flow characteristic, and the third flow characteristic includes the statistical information generated when the first device forwards the flow according to the third congestion control parameter in the third period;
  • the network device obtains the third return value according to the third traffic characteristic; if the third return value is greater than the second return value, the network device continues to use the third step to modify the third congestion control parameter forward to obtain the fourth congestion control parameter,
  • the network device obtains a fourth traffic characteristic, and the fourth traffic characteristic includes statistical information generated when the first device forwards traffic according to the fourth congestion control parameter in the fourth period; the network device obtains a fourth return value according to the fourth traffic characteristic , And so on, until the T+1 reward value is less than the T reward value, the network device uses the first step, the second step, and the sum of
  • step 308 if the second reward value is less than the first reward value, the network device uses the second step size to reversely modify the first congestion control parameter to obtain the third congestion control parameter.
  • the network device uses the second step size to reversely modify the first congestion control parameter to obtain the third congestion control parameter.
  • the reverse modification is just to distinguish it from the forward modification described above. It does not mean that the first congestion control parameter can only be reduced in the reverse direction.
  • the first congestion control parameter is 100
  • the second step is 20
  • the network device After using the second step to reversely modify the first congestion control parameter, the third congestion control parameter can be 80 or 120.
  • the third congestion control parameter is 80 is taken as an example for description.
  • the second step length can be equal to the first step length, for example, the first step length is 20%, and the second step length can also be 20%.
  • step 309 the network device obtains a third traffic characteristic, where the third traffic characteristic includes statistical information generated when the first device forwards traffic according to the third congestion control parameter in the third period; the network device according to the third traffic characteristic, Obtain the third return value.
  • the network device uses the first step to reversely modify the first congestion control parameter. After obtaining the third congestion control parameter, the network device obtains the third flow characteristic.
  • the third flow characteristic includes the first device according to the third congestion control parameter in the third period.
  • the congestion control parameter generates statistical information when forwarding traffic.
  • the third traffic feature can specifically be the statistical information of the first device, or the statistical information of the port of the first device, or the statistical information of the queue of the first device; the network device obtains the third After the traffic characteristics are performed, the network device may obtain a third reward value according to the acquired third traffic characteristics, and the algorithm of the third reward value is similar to the algorithm of the first reward value in step 302.
  • step 310 the network device confirms whether the third reward value is greater than the first reward value.
  • the network device After the network device obtains the third reward value and the first reward value, the network device confirms whether the third reward value is greater than the first reward value.
  • step 311 if the third reward value is greater than the first reward value, the network device uses the second step size to modify the first formula to obtain a second formula, and uses the second formula to obtain a new congestion control parameter.
  • the network device uses the second step size to modify the first formula to obtain the second formula, and uses the second formula to obtain the congestion control parameter.
  • the second formula may be the following algorithm:
  • Q is the congestion control parameter
  • A is the rate
  • B is the set delay
  • C2 is the second step of the percentage value.
  • the network device uses the rate of the first device in the second cycle to set the delay and the first step length.
  • the congestion control parameter can be obtained through the second formula.
  • the congestion control parameter is regarded as the first congestion control parameter, and the second formula is regarded as For the first formula, return to step 302.
  • the network device determines that the third reward value is greater than the first reward value, the network device does not need to modify the first formula in the first step to obtain the second formula, but uses the third step to continue the reverse modification.
  • the third congestion control parameter obtains the fourth congestion control parameter, the network device obtains the fourth flow characteristic, and the fourth flow characteristic includes the statistical information generated when the first device forwards the flow according to the fourth congestion control parameter in the fourth period;
  • the network device obtains the fourth return value according to the fourth traffic characteristic; if the fourth return value is greater than the third return value, the network device continues to use the fourth step to reversely modify the fourth congestion control parameter to obtain the fifth congestion control parameter,
  • the network device obtains a fifth traffic characteristic, and the fifth traffic characteristic includes statistical information generated when the first device forwards traffic according to the fifth congestion control parameter in the fifth period; the network device obtains a fifth return value according to the fifth traffic characteristic , And so on, until the T+1th reward value is less than the Tth reward value, the network device will take the first step, the second step
  • step 312 if the third reward value is less than the first reward value, the network device uses the third step size to modify the first congestion control parameter forward to obtain the fourth congestion control parameter.
  • the network device uses the third step size to forward modify the first congestion control parameter to obtain the fourth congestion control parameter.
  • the third step length is greater than the first step length, because after the network device modifies the first congestion control parameter forward and reversely, the second reward value and the third reward value obtained are both smaller than the first reward value, then There is a probability of falling into a local optimum.
  • the first congestion control parameter is 100
  • the second congestion control parameter is 120
  • the second congestion control parameter is 80.
  • the probability is between 80 and 120.
  • the optimal solution is around 100.
  • the third step length should be greater than the first step length.
  • step 313 the network device obtains a fourth flow characteristic, where the fourth flow characteristic includes statistical information generated when the first device forwards traffic according to the fourth congestion control parameter in the fourth cycle; the network device obtains the fourth flow characteristic according to the fourth flow characteristic, Obtain the fourth return value.
  • the network device uses the third step to modify the first congestion control parameter in the forward direction. After obtaining the fourth congestion control parameter, the network device obtains the fourth flow characteristic.
  • the fourth flow characteristic includes the first device in the fourth cycle according to the fourth Congestion control parameters generate statistical information when forwarding traffic.
  • the fourth traffic feature can specifically be statistical information of the first device, or statistical information of the port of the first device, or statistical information of the queue of the first device; the network device obtains the fourth After the traffic characteristics are performed, the network device can obtain a fourth reward value according to the acquired fourth traffic characteristics, and the algorithm of the fourth reward value is similar to the algorithm of the first reward value in step 302.
  • step 314 the network device confirms whether the fourth reward value is greater than the first reward value.
  • the network device After the network device obtains the fourth reward value and the first reward value, the network device confirms whether the fourth reward value is greater than the first reward value.
  • step 315 if the fourth reward value is greater than the first reward value, the network device uses the third step size to modify the first formula to obtain a second formula, and uses the second formula to obtain a new congestion control parameter.
  • the network device uses the third step size to modify the first formula to obtain the second formula, and uses the second formula to obtain the new congestion control parameter.
  • the second formula may be the following algorithm:
  • Q is the congestion control parameter
  • A is the rate
  • B is the set delay
  • C3 is the third step of the percentage value.
  • the network device uses the rate of the first device in the fourth cycle to set the delay and the first step length.
  • the congestion control parameter can be obtained through the second formula.
  • the congestion control parameter is regarded as the first congestion control parameter, and the second formula is regarded as For the first formula, return to step 302.
  • the network device determines that the fourth reward value is greater than the first reward value, the network device does not need to modify the first formula with the third step length to obtain the step of obtaining the second formula, but uses the fourth step to continue forward modification
  • a fourth congestion control parameter a fifth congestion control parameter is obtained, a network device obtains a fifth flow characteristic, and the fifth flow characteristic includes statistical information generated when the first device forwards traffic according to the fifth congestion control parameter in the fifth cycle;
  • the network device obtains the fifth return value according to the fifth traffic characteristic; if the fifth return value is greater than the fourth return value, the network device continues to use the fifth step to modify the fifth congestion control parameter forward to obtain the sixth congestion control parameter,
  • the network device obtains a sixth traffic characteristic, and the sixth traffic characteristic includes statistical information generated when the first device forwards traffic according to the sixth congestion control parameter in the sixth period; the network device obtains a sixth return value according to the sixth traffic characteristic , And so on, until the T+1 reward value is less than the T reward value, the network device will take the first step, the
  • step 316 if the fourth reward value is less than the first reward value, the network device uses the fourth step to reversely modify the first congestion control parameter to obtain the fifth congestion control parameter.
  • the network device uses the fourth step to reversely modify the first congestion control parameter to obtain the fifth congestion control parameter.
  • the fourth step size is greater than the second step size, because after the network device modifies the first congestion control parameter forward and reversely, the second reward value and the third reward value obtained are both less than the first reward value, then There is a probability of falling into a local optimum.
  • the first congestion control parameter is 100
  • the second congestion control parameter is 120
  • the second congestion control parameter is 80.
  • the probability is between 80 and 120.
  • the optimal solution is around 100.
  • the fourth step length should be greater than the second step length.
  • step 317 the network device obtains a fifth flow characteristic, where the fifth flow characteristic includes statistical information generated when the first device forwards traffic according to the fifth congestion control parameter in the fifth period; the network device obtains the fifth flow characteristic according to the fifth flow characteristic, Obtain the fifth return value.
  • the network device uses the fourth step to reversely modify the first congestion control parameter. After obtaining the fifth congestion control parameter, the network device obtains the fifth flow characteristic.
  • the fifth flow characteristic includes the first device according to the fifth Congestion control parameters generate statistical information when forwarding traffic.
  • the fifth traffic feature can specifically be statistical information of the first device, or statistical information of the port of the first device, or statistical information of the queue of the first device; the network device obtains the fifth After the traffic characteristic is performed, the network device can obtain a fifth reward value according to the acquired fifth traffic characteristic, and the algorithm of the fifth reward value is similar to the algorithm of the first reward value in step 302.
  • step 318 the network device confirms whether the fifth reward value is greater than the first reward value.
  • the network device After the network device obtains the fifth reward value and the first reward value, the network device confirms whether the fifth reward value is greater than the first reward value.
  • step 319 if the fifth reward value is greater than the first reward value, the network device uses the fourth step size to modify the first formula to obtain the second formula, and uses the second formula to obtain the new congestion control parameter.
  • the network device uses the fourth step size to modify the first formula to obtain the second formula, and uses the second formula to obtain the congestion control parameter.
  • the second formula may be the following algorithm:
  • Q is the congestion control parameter
  • A is the rate
  • B is the set delay
  • C4 is the fourth step of the percentage value.
  • the network device uses the rate of the first device in the fifth cycle to set the delay and the first step length.
  • the congestion control parameter can be obtained through the second formula.
  • the congestion control parameter is regarded as the first congestion control parameter, and the second formula is regarded as For the first formula, return to step 302.
  • the network device determines that the fifth reward value is greater than the first reward value, the network device does not need to modify the first formula with the fourth step length to obtain the step of obtaining the second formula, but uses the fifth step length to continue the reverse modification
  • a fifth congestion control parameter, a sixth congestion control parameter is obtained, a network device obtains a sixth flow characteristic, and the sixth flow characteristic includes statistical information generated when the first device forwards traffic according to the sixth congestion control parameter in the sixth cycle;
  • the network device obtains the sixth return value according to the sixth traffic characteristic; if the sixth return value is greater than the fifth return value, the network device continues to use the sixth step to reversely modify the sixth congestion control parameter to obtain the seventh congestion control parameter,
  • the network device obtains the seventh traffic characteristic, and the seventh traffic characteristic includes the statistical information generated when the first device forwards traffic according to the seventh congestion control parameter in the seventh cycle; the network device obtains the seventh return value according to the seventh traffic characteristic , And so on, until the T+1th reward value is less than the Tth reward value, the network device will take
  • step 320 if the fifth return value is greater than the first return value, there is a F% probability that the network device will use the third step size as the first step size and the fourth step size as the second step size, and return to step 312 , The network device has a probability of G% and returns to step 301.
  • step 312 the network device has a hundred percent probability. The probability of dividing G is returned to step 301.
  • F plus G equals 100.
  • the value of F decays with the number of times of returning to step 312, for example, the first time F is equal to 50, after step 320, it returns to step 312, from step 312 to step 320, the value of F becomes 40, then In this determination, there is a 40% probability that it will return to step 312.
  • the first congestion control parameter is multiple parameters.
  • the first congestion control parameter includes three parameters, namely the lower line, the upper line, and the maximum marking probability.
  • FIG. 4 is another flow diagram of the congestion control method provided by this application.
  • step 401 the network device obtains a first congestion control parameter.
  • the network device is configured with the first congestion control rule.
  • the first congestion control rule is the AI model.
  • the network device can input the collected initial flow characteristics into the AI model to obtain the AI model
  • the output first congestion control parameter, the first congestion control parameter includes the first downline, the first upline, and the first maximum marking probability.
  • the network device can use the pre-set traffic characteristic as the initial traffic characteristic, or do not acquire the first congestion control parameter in this cycle, and wait for the next cycle to acquire the traffic of the current cycle.
  • the characteristic is used as the initial flow characteristic, and the first congestion control parameter is obtained by using the flow characteristic of the current cycle to control the flow of the first device in the next cycle.
  • the first congestion control rule can also be a formula.
  • a formula can be selected as the first congestion control rule.
  • this embodiment takes the first congestion control rule being an AI model as an example for description.
  • the network device may be the first device, or may be a device other than the first device.
  • the network device obtains a first traffic characteristic
  • the first traffic characteristic includes the statistical information generated when the first device forwards traffic according to the first congestion control parameter in the first period, and the network device obtains the first traffic characteristic according to the first traffic characteristic.
  • a return value A return value.
  • the network device obtains the first operating value and the second operating value of the first device, and determines whether the difference between the first operating value and the second operating value is less than a target threshold.
  • Step 402 and step 403 are similar to step 302 and step 303 in the aforementioned FIG. 3.
  • step 404 if the difference between the first operating value and the second operating value is less than the target threshold, the network device uses the first step to forwardly modify the first downline to obtain the second downline.
  • the network device uses the first long step to modify the first downline to obtain the second downline, and the network device transfers the second downline to the second downline.
  • the line, the first water line, and the first maximum marking probability are used as the second congestion control parameters, and other descriptions are similar to step 304 in FIG. 3 described above.
  • step 405 the network device obtains a second traffic characteristic, and the second traffic characteristic includes statistical information generated when the first device forwards traffic according to the second congestion control parameter; the network device obtains a second return value according to the second traffic characteristic .
  • Step 405 is similar to step 305 in FIG. 3 described above.
  • step 406 the network device positively modifies the first upper waterline by using the second step size to obtain the second forward waterline.
  • the network device uses the second step to modify the first upstream waterline forward to obtain the second forward waterline.
  • the network device uses the target downstream waterline, the second forward waterline, and the first maximum marking probability as the third congestion control parameter.
  • the target downline includes the first downline or the second positive downline. If the second return value is greater than the first downline, the target downline includes the second downline, and if the second return value is greater than the first downline ,
  • the target launch line includes the first launch line.
  • the network device can use the second step to modify the second forward downward pipeline to obtain the third forward downward pipeline, and the network device can set the third forward downward pipeline ,
  • the first water line and the first maximum marking probability are used as the third congestion control parameter, and the network device obtains the third traffic characteristic, and the third traffic characteristic includes that the first device forwards the traffic according to the third congestion control parameter in the third period
  • the network device using the first long forward modification is inferior to the first downward pipeline, and the network device uses the second positive downward pipeline.
  • the AI model inputs the same initial flow characteristics, reduce the probability that the AI model will output the second positive downward pipeline.
  • the initial flow characteristics are 50
  • the AI model outputs the first downline 10, the first upline 60, and the probability of the first maximum marking probability 50 is 30%.
  • the AI model outputs the second positive downline 12, the first upline 30, and the first The probability of the maximum marking probability of 50 is 7%.
  • the network equipment modifies the AI model.
  • the AI model When the AI model inputs the initial flow characteristics of 50, the AI model outputs the second positive downward waterline 12, the first upper waterline 30, and the first maximum marking probability 50 The probability is 3%. Therefore, the next time the network device obtains the initial traffic characteristics 50 and uses the AI model to reason about the congestion control parameters, the AI model outputs the second forward downward pipeline 12, the first upward pipeline 30, and the probability of the first maximum marking probability 50 will decrease. .
  • the network device uses the second step to modify the first waterline forward and the network device uses the second step to modify the second positive downward waterline.
  • There is a second step in the second step but this second step is only a convenience
  • the description does not limit the values of the two second step lengths to be equal, similar, in this embodiment, even if the same step length appears, it is necessary to determine whether the object of step length modification is the same, only the step length is the same, and Only when the target of step modification is the same can it be determined that the values of the two steps are the same.
  • the network device uses the second step to positively modify the first waterline.
  • the modified object is the first waterline, and the network device uses the first waterline.
  • Two-step forward modification of the second positive downward waterline The object of modification is the second forward downward waterline, so the values of these two steps may be different.
  • the network device obtains a third traffic characteristic, and the third traffic characteristic includes statistical information generated when the first device forwards traffic according to the third congestion control parameter; the network device obtains a third return value according to the third traffic characteristic .
  • the network device obtains a third traffic characteristic, where the third traffic characteristic includes statistical information generated when the first device forwards traffic according to the third congestion control parameter; the network device obtains a third return value according to the third traffic characteristic.
  • step 408 the network device uses the third step size to forwardly modify the first maximum marking probability to obtain the second forward maximum marking probability.
  • the network equipment uses the third step to modify the first maximum marking probability forward to obtain the second forward maximum marking probability; the network equipment uses the target launch line, the target upload line and the second maximum marking probability as the fourth congestion control parameter, the target The downline includes the first downline or the second positive downline. If the second return value is greater than the first downline, the target downline includes the second downline.
  • the target lower line includes the first lower line
  • the target upper line includes the first upper line or the second positive upper line
  • the third return value is greater than the first return value
  • the third return value is greater than the second return value
  • the target waterline includes the second positive waterline, otherwise the target waterline includes the first waterline.
  • the network device may use the third step to modify the second forward waterline in a positive direction.
  • the network device obtains a fourth flow characteristic, where the fourth flow characteristic includes statistical information generated when the first device forwards the flow according to the fourth congestion control parameter; the network device obtains a fourth return value according to the fourth flow characteristic .
  • the network device obtains a fourth flow characteristic, where the fourth flow characteristic includes statistical information generated when the first device forwards the flow according to the fourth congestion control parameter; the network device obtains a fourth return value according to the fourth flow characteristic.
  • step 410 the network device uses the fourth step size to reversely modify the first downline to obtain the second reverse downline.
  • the network equipment uses the fourth step to reversely modify the first offline to obtain the second reverse offline.
  • the network equipment uses the second reverse offline, the target upstream, and the target maximum marking probability as the fifth congestion control parameter,
  • the target upper waterline includes the first upper waterline or the second positive upper waterline.
  • the target upper waterline includes the second positive upper waterline Water line, otherwise the target water line includes the first water line, the target maximum marking probability includes the first maximum marking probability or the second forward maximum marking probability, if the fourth reward value is greater than the third reward value, and the fourth reward value If it is greater than the second reward value and the fourth reward value is greater than the first reward value, the target maximum marking probability includes the second forward maximum marking probability, otherwise the target maximum marking probability includes the first maximum marking probability.
  • the network device can use the fourth step size
  • the second forward maximum mark probability is modified in the forward direction.
  • the network device obtains a fifth flow characteristic, and the fifth flow characteristic includes statistical information generated when the first device forwards the flow according to the fifth congestion control parameter; the network device obtains a fifth return value according to the fifth flow characteristic .
  • the network device obtains a fifth flow characteristic, where the fifth flow characteristic includes statistical information generated when the first device forwards the flow according to the fifth congestion control parameter; the network device obtains a fifth return value according to the fifth flow characteristic.
  • step 412 the network device uses the fifth step to reversely modify the first water line to obtain the second reverse water line.
  • the network device uses the fifth step to reversely modify the first upper waterline to obtain the second reverse upper waterline.
  • the network device uses the reverse target offline, the second reverse upper waterline, and the maximum marking probability of the target as the sixth congestion control parameter ,
  • the reverse target launch line includes the target launch line or the second reverse launch line, if the fifth return value is greater than the fourth return value, and the fifth return value is greater than the third return value, and the fifth return value is greater than the second return value ,
  • the fifth reward value is greater than the first reward value
  • the reverse target pipeline includes the second reverse pipeline, otherwise the reverse target pipeline includes the target pipeline, and the target maximum marking probability includes the first maximum marking probability or the second Maximum positive mark probability.
  • the target maximum mark probability includes the second positive maximum Marking probability, otherwise the target maximum marking probability includes the first maximum marking probability.
  • step 406 if the fifth reward value is greater than the fourth reward value, and the fifth reward value is greater than the third reward value, and the fifth reward value is greater than the second reward value, and the fifth reward value is greater than the first reward value.
  • the network device can use the fifth step to reversely modify the second reverse pipeline.
  • the network device obtains a sixth traffic characteristic, where the sixth traffic characteristic includes statistical information generated when the first device forwards traffic according to the sixth congestion control parameter; the network device obtains a sixth return value according to the sixth traffic characteristic .
  • the network device obtains a sixth traffic characteristic, where the sixth traffic characteristic includes statistical information generated when the first device forwards traffic according to the sixth congestion control parameter; the network device obtains a sixth return value according to the sixth traffic characteristic.
  • step 414 the network device uses the sixth step size to reversely modify the first maximum marking probability to obtain the second reverse maximum marking probability.
  • the network equipment uses the sixth step to reversely modify the first maximum marking probability to obtain the second reverse maximum marking probability.
  • the network equipment takes the reverse target offline, the reverse target upstream, and the second reverse maximum marking probability as the first Seven congestion control parameters, the reverse target launch line includes the target launch line or the second reverse launch line, if the fifth return value is greater than the fourth return value, and the fifth return value is greater than the third return value, and the fifth return value is greater than The second return value, and the fifth return value is greater than the first return value, then the reverse target lower line includes the second reverse lower line, otherwise the reverse target lower line includes the target lower line, and the reverse target upper line includes the target upper line Water line or second reverse upper water line, if the sixth reward value is greater than the fifth reward value, and the sixth reward value is greater than the fourth reward value, and the sixth reward value is greater than the third reward value, and the sixth reward value is greater than the first Two return values, and the sixth return value is greater than the first return value, then the reverse target upper waterline includes the second reverse upper waterline,
  • the network device can use the sixth step to reversely modify the second reverse upper waterline.
  • the network device obtains the seventh traffic characteristic, the seventh traffic characteristic includes the statistical information generated when the first device forwards the traffic according to the seventh congestion control parameter; the network device obtains the seventh return value according to the seventh traffic characteristic .
  • the network device obtains the seventh traffic characteristic, where the seventh traffic characteristic includes the statistical information generated when the first device forwards the traffic according to the seventh congestion control parameter; the network device obtains the seventh return value according to the seventh traffic characteristic.
  • step 416 the network device obtains the new initial flow characteristics, and the network device inputs the new initial flow characteristics into the AI model for inference, and obtains the initial congestion control parameters.
  • the network device obtains new initial traffic characteristics, and the network device inputs the new initial traffic characteristics into the AI model for inference, and obtains initial congestion control parameters.
  • the initial congestion control parameters are used to allow the first device to control the forwarded traffic according to the initial congestion control parameters.
  • the network device does not need to use the step size to modify the initial congestion control parameter, and directly uses the initial congestion control parameter as a parameter for the first device to control traffic forwarding.
  • the network device uses the seventh congestion control parameter to modify the AI model, and the network device uses the modified AI model to perform Reasoning to obtain the initial congestion control parameters.
  • the seventh congestion control parameter includes reverse target launch line, reverse target launch line, and reverse target maximum marking probability.
  • the reverse target launch line includes the target launch line or the second reverse launch line.
  • the reverse target pipeline includes the second reverse pipeline, otherwise the reverse target pipeline includes the target pipeline, and the target pipeline includes the first pipeline or the second forward pipeline
  • the lower line if the second return value is greater than the first return value, the target lower line includes the second forward lower line, and if the second return value is greater than the first return value, the target lower line includes the first lower line; reverse The target upper waterline includes the target upper waterline or the second reverse upper waterline.
  • the reverse target upper waterline includes the second reverse upper waterline, otherwise the reverse target upper waterline includes the target upper waterline,
  • the target upper waterline includes the first upper waterline or the second positive upper waterline.
  • the target upper waterline includes the second positive upper waterline Waterline, otherwise the target upper waterline includes the first upper waterline; the maximum reverse target marking probability includes the target maximum marking probability or the second reverse maximum marking probability; if the seventh reward value is greater than the sixth reward value, and the seventh reward Value is greater than the fifth reward value, and the seventh reward value is greater than the fourth reward value, and the seventh reward value is greater than the third reward value, and the seventh reward value is greater than the second reward value, and the seventh reward value is greater than the first reward value ,
  • the reverse target maximum marking probability includes the second reverse maximum marking probability, otherwise the reverse target maximum marking probability includes the target maximum marking probability, and the target maximum marking probability includes the first maximum marking probability or the second forward maximum marking probability, if If the fourth reward value is greater than the third reward value, and the fourth reward value is greater than the second reward value, and the fourth reward value is greater than the first reward value, then the target maximum marking probability includes the second positive maximum marking
  • the AI model when the AI model inputs the same initial traffic characteristics, increase the probability that the AI model will output the seventh congestion control parameter.
  • the initial traffic characteristics are 50
  • the AI model input the initial traffic characteristics 50 to the AI model, and the AI model outputs the first one.
  • Waterline 10 the first upper waterline 60
  • the probability of the first maximum marking probability 50 is 30%
  • the AI model outputs the seventh congestion control parameter: the second forward downward pipeline 12, the second reverse downward pipeline 27, and the second The probability of the maximum forward marking probability of 60 is 7%.
  • the network equipment modifies the AI model.
  • the AI model inputs the initial traffic characteristics of 50
  • the AI model outputs the seventh congestion control parameter: the second forward downward pipeline is 12, and the second reverse launch Line 27, the probability of the second maximum marking probability 60 in the positive direction is 50%.
  • the AI model outputs the seventh congestion control parameter: the second forward downward pipeline 12, the second reverse downward pipeline 27, and the second forward pipeline.
  • the probability of reaching the maximum mark probability of 60 will increase, from 7% to 50%.
  • step 406 if the seventh reward value is greater than the sixth reward value, and the seventh reward value is greater than the fifth reward value, and the seventh reward value is greater than the fourth reward value, and the seventh reward value is greater than the third reward value. If the reward value is greater than the second reward value, and the seventh reward value is greater than the first reward value, the network device can use the seventh step to reversely modify the second reverse maximum marking probability.
  • a single device in addition to the network device, can use the congestion control method to control the flow forwarded by the device, or multiple devices can use the congestion control method to control the flow forwarded by the device.
  • the situation of a single device is described above, and the situation of multiple devices is described below.
  • FIG. 5 is a schematic flowchart of another embodiment of the congestion control method provided by this application.
  • step 501 the network device obtains the first congestion control parameter and the second congestion control parameter.
  • the network device is configured with the first congestion control rule. When the network device needs to control the flow of the first device, the network device can use the first congestion control rule to obtain the first congestion according to the first initial flow characteristics collected from the first device. control parameter.
  • the network device is also configured with a second congestion control rule. When the network device needs to control the flow of the second device, the network device can use the second congestion control rule to obtain the second congestion control rule according to the second initial flow characteristics collected from the second device. Congestion control parameters.
  • the network device can use the pre-set flow characteristic as the initial flow characteristic, or not obtain the first congestion control parameter and the second congestion control parameter in this cycle, and wait for the next Periodically obtain the flow characteristics of the current period as the initial flow characteristics, use the flow characteristics of the first device in the current period to obtain the first congestion control parameter to control the flow of the first device in the next period, and use the flow characteristics of the second device in the current period to obtain the first The second congestion control parameter controls the flow of the second device in the next cycle.
  • the first congestion control rule can be an AI model, and the first congestion control rule can also be a formula. In practical applications, one can be selected as the first congestion control rule.
  • step 502 the network device obtains the first flow characteristic and the second flow characteristic, the network device obtains a first reward value according to the first flow characteristic, and obtains a second reward value according to the second flow characteristic.
  • the network device obtains the first traffic characteristic and the second traffic characteristic.
  • the first traffic characteristic includes the statistical information generated when the first device forwards traffic according to the first congestion control parameter in the first period.
  • the network device obtains the first traffic characteristic according to the first traffic characteristic.
  • a return value, the second traffic characteristic includes statistical information generated when the second device forwards traffic according to the second congestion control parameter in the first period, and the network device obtains the second return value according to the second traffic characteristic.
  • the method for obtaining the reward value is similar to the method for obtaining the reward value in step 302 of FIG. 3 described above, and the details are not repeated here.
  • the network device obtains the first operating value and the second operating value of the first device, obtains the third operating value and the fourth operating value of the second device, and determines whether the average value of the difference is less than the target threshold.
  • the network device obtains the first operating value and the second operating value of the first device, obtains the third operating value and the fourth operating value of the second device, and determines whether the average value of the difference is less than the target threshold.
  • the average value of the difference refers to the first The average value of the difference value and the second difference value.
  • the first difference value refers to the difference value between the first operation value and the second operation value
  • the second difference value is the difference value between the third operation value and the fourth operation value.
  • the network device may adjust the initial congestion control parameter only through the first congestion control rule, and adjusting the initial congestion control rule only through the first congestion control rule means that the network device does not use the step size to adjust the network device according to the first congestion control rule.
  • the network device directly uses the congestion control parameter obtained by the first congestion control rule as the parameter for the first device to control traffic forwarding; the network device uses the first step to modify the first congestion control parameter to obtain the second congestion control parameter.
  • the network device may count the number of consecutive adjustments of the initial congestion control parameter by the network device only through the first congestion control rule.
  • step 504 if the mean value of the difference is less than the target threshold, the network device uses the first step to forwardly modify the first congestion control parameter to obtain the third congestion control parameter, and the network device uses the second step to forwardly modify the second congestion control parameter. Congestion control parameters to obtain the fourth congestion control parameter.
  • the network device uses the first step to forwardly modify the first congestion control parameter to obtain the third congestion control parameter, and the network device uses the second step to forwardly modify the second congestion control parameter to obtain The fourth congestion control parameter.
  • the forward modification is only to distinguish it from the reverse modification described later. It does not mean that the first congestion control parameter can only be added forward.
  • the first congestion control parameter is 100
  • the first step length is 20, and the network device
  • the second congestion control parameter can be 80 or 120.
  • the second congestion control parameter is 120 is taken as an example for description.
  • the first step length and the second step length are percentage values.
  • the target threshold can be adjusted according to the size of the network fluctuation of the first device.
  • the network device reduces the target threshold.
  • the network fluctuates greatly, that is, the traffic of the first device changes greatly, and the network device increases the target threshold.
  • the network device may not obtain the first operating value and the second operating value of the first device, and the network device may Determine whether the number of consecutive adjustments is greater than the set threshold N.
  • the network device executes the first long forward modification of the first congestion control parameter to obtain the second congestion control Parameter steps.
  • the network device when the network device counts the number of consecutive adjustments of the initial congestion control parameter by the network device only through the first congestion control rule, the network device obtains the first operating value, the second operating value, the third operating value, and the fourth operating value. At the operating value, as long as one condition is met, the network device executes the step of using the first step to forwardly modify the first congestion control parameter to obtain the second congestion control parameter, which includes the following situations:
  • the network device executes the first step of forward modification of the first congestion control parameter , The step of obtaining the second congestion control parameter.
  • the network device executes the first step of forward modification of the first congestion control parameter , The step of obtaining the second congestion control parameter.
  • the network device executes the first step of forward modification of the first congestion control parameter , The step of obtaining the second congestion control parameter.
  • step 503 may not be performed.
  • the network device does not need to confirm whether the mean value of the difference is less than the target threshold, and directly uses the first long forward Modify the first congestion control parameter to obtain the second congestion control parameter, and use the second step size to modify the second congestion control parameter forward to obtain the fourth congestion control parameter.
  • step 505 the network device obtains the third flow characteristic and the fourth flow characteristic, the network device obtains the third reward value according to the third flow characteristic, and the network device obtains the fourth reward value according to the fourth flow characteristic.
  • the network device uses the first step to positively modify the first congestion control parameter.
  • the network device can obtain a third flow characteristic.
  • the third flow characteristic includes the first device according to the first congestion control parameter in the second period.
  • Three congestion control parameters are the statistical information generated when the traffic is forwarded; the third traffic feature can specifically be the statistical information of the first device, or the statistical information of the port of the first device, or the statistical information of the queue of the first device; the network device obtains the first device’s statistical information.
  • the network device can obtain a third reward value according to the acquired third flow characteristic, and the algorithm of the third reward value is similar to the algorithm of the first reward value in step 302 in FIG. 3 described above.
  • the third flow characteristic may be a value of the first device at a certain moment in the second period, or a processed value.
  • the third flow characteristic may be an average value of the first device in the second period.
  • the network device uses the second step to modify the second congestion control parameter forward. After obtaining the fourth congestion control parameter, the network device can obtain the fourth flow characteristic.
  • the fourth flow characteristic includes the second device according to the second congestion control parameter in the second period.
  • the fourth congestion control parameter is the statistical information generated when the traffic is forwarded; the fourth traffic feature may specifically be the statistical information of the second device, or the statistical information of the port of the second device, or the statistical information of the queue of the second device; the network device obtains the first
  • the network device can obtain a fourth reward value according to the acquired fourth flow characteristic.
  • the algorithm of the fourth reward value is similar to the algorithm of the first reward value in step 302 in FIG. 3 described above.
  • the fourth flow characteristic may be the value of the second device at a certain moment in the second period, or a processed value. For example, the fourth flow characteristic may be the average value of the second device in the second period.
  • the network device when the network device is the first device, the network device directly issues the third congestion control parameter to the forwarding chip, so that the forwarding chip controls the traffic forwarded by the first device in the second period, and the network device sends the first device to the second device.
  • the fourth congestion control parameter is used to allow the second device to control the forwarded traffic according to the fourth congestion control parameter.
  • the network device when the network device is the second device, the network device directly issues the fourth congestion control parameter to the forwarding chip, so that the forwarding chip controls the traffic forwarded by the second device in the second period, and the network device sends the first device to the first device.
  • the third congestion control parameter is used to allow the third device to control the forwarded traffic according to the third congestion control parameter.
  • the network device when the network device is a device other than the first device and the second device, after the network device obtains the third congestion control parameter and the fourth congestion control parameter, the network device sends the third congestion control parameter to the first device
  • the parameter is used to allow the first device to use the third congestion control parameter to control the traffic forwarded by the first device in the second cycle.
  • the network device will also send the fourth congestion control parameter to the second device for the second device to use the Four congestion control parameters are used to control the traffic forwarded by the second device in the second cycle.
  • step 506 the network device confirms whether the sum of the third reward value and the fourth reward value is greater than the sum of the first reward value and the second reward value.
  • step 507 if the sum of the third reward value and the fourth reward value is greater than the sum of the first reward value and the second reward value, the network device uses the first step to modify the first congestion control rule to obtain the third congestion control Rule, use the second step to modify the second congestion control rule to obtain the fourth congestion control rule.
  • the network device determines that the sum of the third reward value and the fourth reward value is greater than the sum of the first reward value and the second reward value, the network device does not need to modify the first congestion control rule in the first step to obtain the first congestion control rule.
  • Three congestion control rules use the second step size to modify the second congestion control rule to obtain the fourth congestion control rule step, but use the third step size to continue to modify the third congestion control parameter forward to obtain the fifth congestion control parameter, Use the fourth step to continue to modify the fourth congestion control parameter in the forward direction to obtain the sixth congestion control parameter;
  • the network device obtains the fifth flow characteristic, and the fifth flow characteristic includes the first device according to the fifth congestion control in the third cycle Parameter forwarding traffic;
  • the network device obtains the sixth traffic characteristic, and the sixth traffic characteristic includes the statistical information generated when the second device forwards the traffic according to the sixth congestion control parameter in the third period;
  • the network device obtains the statistics according to the sixth congestion control parameter;
  • Five flow characteristics obtain the fifth return value; the network device obtains the sixth return value according to the
  • the network equipment uses the first step length, the third step length, and the sum of all the step lengths to the T-1 step length as the first step length, and uses the first step length to modify the first congestion control Rule, obtain the third congestion control rule
  • the network device uses the second step, the fourth step, and the sum of all the steps to the T-th step as the second step, and uses the second step to modify the second congestion control Rule to obtain the fourth congestion control rule.
  • step 508 if the sum of the third reward value and the fourth reward value is less than the sum of the first reward value and the second reward value, the network device uses the fifth step to reversely modify the first congestion control parameter to obtain the fifth Congestion control parameters, use the sixth step size to modify the first congestion control parameter to obtain the sixth congestion control parameter.
  • step 509 the network device obtains the fifth flow characteristic and the sixth flow characteristic, the network device obtains the fifth reward value according to the fifth flow characteristic, and the network device obtains the sixth reward value according to the sixth flow characteristic.
  • the network device uses the third step to reversely modify the first congestion control parameter.
  • the network device can obtain the fifth flow characteristic.
  • the fifth flow characteristic includes the first device according to the first congestion control parameter in the third period.
  • the fifth congestion control parameter is the statistical information generated when the traffic is forwarded; the fifth traffic feature may specifically be the statistical information of the first device, or the statistical information of the port of the first device, or the statistical information of the queue of the first device; the network device obtains the first
  • the network device can obtain a fifth reward value according to the acquired fifth flow characteristic, and the algorithm of the fifth reward value is similar to the algorithm of the first reward value in step 302 in FIG. 3 described above.
  • the fifth flow characteristic may be the value of the first device at a certain moment in the third period, or a processed value.
  • the fifth flow characteristic may be the average value of the first device in the third period.
  • the network device uses the fourth step to reversely modify the second congestion control parameter.
  • the network device can obtain the sixth flow characteristic.
  • the sixth flow characteristic includes the second device according to the second congestion control parameter in the third cycle.
  • the statistical information generated when the congestion control parameter forwards the traffic can specifically be the statistical information of the second device, or the statistical information of the port of the second device, or the statistical information of the queue of the second device; the network device obtains the first
  • the network device can obtain a sixth reward value according to the acquired sixth traffic characteristic, and the algorithm of the sixth reward value is similar to the algorithm of the first reward value in step 302 in FIG. 3 described above.
  • the sixth flow characteristic may be the value of the second device at a certain moment in the third period, or a processed value.
  • the sixth flow characteristic may be the average value of the second device in the third period.
  • step 510 if the sum of the fifth reward value and the sixth reward value is greater than the sum of the first reward value and the second reward value, the network device uses the fifth step to modify the first congestion control rule to obtain the third congestion control Rule, use the sixth step to modify the second congestion control rule to obtain the fourth congestion control rule.
  • the network device determines that the sum of the fifth reward value and the sixth reward value is greater than the sum of the first reward value and the second reward value, the network device does not need to modify the fifth step length first.
  • a congestion control rule, the third congestion control rule is obtained, the second congestion control rule is modified by the sixth step, and the fourth congestion control rule is obtained, but the seventh step is used to continue to modify the fifth congestion control parameter in the reverse direction, Obtain the seventh congestion control parameter, continue to modify the sixth congestion control parameter in the reverse direction using the eighth step size, and obtain the eighth congestion control parameter.
  • the congestion control method in the embodiment of the present application is described above, and the congestion control apparatus in the embodiment of the present application is described below.
  • FIG. 6 is a schematic structural diagram of an embodiment of the congestion control apparatus provided by this application.
  • the first acquiring unit 601 is configured to acquire a first traffic characteristic, where the first traffic characteristic includes statistical information generated when the first device forwards traffic according to the first congestion control parameter in the first cycle, and the first congestion control parameter is based on the first congestion control parameter. Obtained by congestion control rules;
  • the second obtaining unit 602 is configured to obtain the first reward value according to the first traffic characteristic
  • the third obtaining unit 603 is configured to modify the first congestion control parameter by using the first step to obtain the second congestion control parameter;
  • the fourth acquiring unit 604 is configured to acquire a second traffic characteristic, where the second traffic characteristic includes statistical information generated when the first device forwards traffic according to the second congestion control parameter in the second period;
  • the fifth obtaining unit 605 is configured to obtain a second return value according to the second flow characteristic
  • the execution unit 606 is configured to execute corresponding processing if the second reward value is greater than the first reward value.
  • the first obtaining unit 601 may obtain the first flow characteristic, and the first flow characteristic includes the first device according to the first congestion in the first period.
  • the first congestion control parameter is obtained according to the first congestion control rule, and the second obtaining unit 602 can obtain the first return value according to the first traffic characteristic.
  • the third obtaining unit 603 can obtain the second congestion control parameter, and the fourth obtaining unit 604 can obtain the second flow characteristic.
  • the second flow characteristic includes the first device according to the second congestion in the second period.
  • the statistical information generated when the control parameter forwards the traffic can obtain the second reward value according to the second traffic characteristic, and if the second reward value is greater than the first reward value, the execution unit 606 executes corresponding processing, wherein ,
  • the third obtaining unit 603 uses the first step to modify the first congestion control parameter, and obtains the second congestion control parameter that can obtain a larger reward value. Because the second reward value is greater than the first reward value, the second congestion control parameter It is better than the first congestion control parameter, so the congestion control device optimizes the inference result of the first congestion control rule, thereby improving the generalization of the first congestion control rule.
  • each unit of the congestion control apparatus is similar to those described in the foregoing embodiment shown in FIG. 2 and will not be repeated here.
  • FIG. 7 is a schematic structural diagram of another embodiment of the congestion control apparatus provided by this application.
  • the congestion control device provided in this application further includes:
  • the first obtaining unit 601 is further configured to obtain the first operating value and the second operating value of the first device.
  • the congestion control device also includes:
  • the determining unit 707 is configured to determine whether the difference between the first operating value and the second operating value is less than a target threshold
  • the third acquiring unit 603 is specifically configured to, if the difference value is less than the target threshold, execute the step of using the first step to modify the first congestion control parameter to obtain the second congestion control parameter.
  • the congestion control device further includes:
  • the adjustment unit 708 is configured to adjust initial congestion control parameters
  • the statistics unit 709 is used to count the number of consecutive adjustments of initial congestion control parameters
  • the third obtaining unit 603 is specifically configured to perform the step of modifying the first congestion control parameter by using the first step length to obtain the second congestion control parameter when the number of consecutive adjustments is greater than the set threshold N;
  • the first congestion control parameter is obtained by continuously adjusting the initial congestion control parameter N times according to the first congestion control rule.
  • the first congestion control rule is a first formula, and the first formula is:
  • Q F(A, B), where Q is the congestion control parameter, A is the rate, B is the set delay, and F(A, B) is the function related to A and B.
  • the fifth obtaining unit 605 is further configured to modify the first formula by using the first step length to obtain the second formula.
  • the fifth obtaining unit 605 is further configured to obtain the third congestion control parameter by using a second formula according to the second traffic characteristic;
  • the third congestion control parameter is used for the first device to control the forwarded traffic.
  • the first step length is a percentage value.
  • the second formula is:
  • the execution unit 606 is specifically configured to use the second step size to forward modify the second congestion control parameter to obtain the third congestion control parameter;
  • the execution unit 606 is specifically configured to obtain a third traffic characteristic, where the third traffic characteristic includes statistical information generated when the first device forwards traffic according to the third congestion control parameter in the third period;
  • the execution unit 606 is specifically configured to obtain the third reward value according to the third traffic characteristic.
  • the third obtaining unit 603 is further configured to, if the third reward value is less than the second reward value, use the third step size to reversely modify the second congestion control parameter to obtain the fourth congestion control parameter;
  • the fourth acquiring unit 604 is further configured to acquire a fourth traffic characteristic, where the fourth traffic characteristic includes statistical information generated when the first device forwards traffic according to the fourth congestion control parameter in the fourth cycle;
  • the fifth obtaining unit 605 is further configured to obtain a fourth reward value according to the fourth traffic characteristic.
  • the congestion control device further includes:
  • the modifying unit 710 is configured to, if the fourth reward value is greater than the second reward value, modify the first congestion control rule by using the third step length and the first step length to obtain the second congestion control rule.
  • the congestion control device further includes:
  • the generating unit 711 is configured to use the second congestion control rule to generate a new congestion control parameter of the first device according to the fourth traffic characteristic, and the new congestion control parameter is used for the first device to control the forwarded traffic.
  • the second step length is greater than the first step length.
  • the first obtaining unit 601 is further configured to obtain a fifth flow characteristic, and the fifth flow characteristic includes statistical information generated when the second device forwards traffic according to the fifth congestion control parameter in the first cycle, and the fifth congestion control parameter Obtained according to the third congestion control rule;
  • the second obtaining unit 602 is further configured to obtain a fifth return value according to the fifth flow characteristic
  • the third obtaining unit 603 is further configured to modify the fifth congestion control parameter by using the fourth step size to obtain the sixth congestion control parameter;
  • the fourth acquiring unit 604 is further configured to acquire a sixth traffic characteristic, where the sixth traffic characteristic includes statistical information generated when the second device forwards traffic according to the sixth congestion control parameter in the second cycle;
  • the fifth obtaining unit 605 is further configured to obtain a sixth return value according to the sixth flow characteristic
  • the execution unit 606 is further configured to execute the corresponding processing if the sum of the sixth reward value and the second reward value is greater than the sum of the fifth reward value and the first reward value.
  • the execution unit 706 is specifically configured to modify the third congestion control rule according to the fourth step size
  • the execution unit 606 is specifically configured to modify the first congestion control rule according to the first step length.
  • the execution unit 706 is specifically configured to use the second step size to modify the second congestion control parameter to obtain the third congestion control parameter;
  • the execution unit 606 is specifically configured to obtain a third traffic characteristic, where the third traffic characteristic includes statistical information generated when the first device forwards traffic according to the third congestion control parameter in the third period;
  • the execution unit 606 is specifically configured to obtain the third reward value according to the third flow characteristic
  • the execution unit 606 is specifically configured to use the fifth step size to modify the sixth congestion control parameter to obtain the seventh congestion control parameter;
  • the execution unit 606 is specifically configured to obtain a seventh traffic characteristic, where the seventh traffic characteristic includes statistical information generated when the second device forwards traffic according to the seventh congestion control parameter in the third period;
  • the execution unit 606 is specifically configured to obtain the seventh reward value according to the seventh flow characteristic.
  • each unit of the congestion control apparatus is similar to those described in the foregoing embodiments shown in FIG. 2 and FIG. 3 and FIG. 4, and will not be repeated here.
  • FIG. 8 is a schematic structural diagram of an embodiment of the congestion control device provided by this application.
  • the congestion control device 800 includes a processor 810, a memory coupled with the processor 810, and a communication interface 830.
  • the congestion control device 800 may be the network device of FIG. 1, the first device or the second device.
  • the processor 810 may be a central processing unit (CPU), a network processor (NP), or a combination of a CPU and an NP.
  • the processor may also be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof.
  • ASIC application-specific integrated circuit
  • PLD programmable logic device
  • the above-mentioned PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a generic array logic (GAL) or any combination thereof.
  • the processor 810 may refer to one processor, or may include multiple processors.
  • the memory may include volatile memory (volatile memory), such as random access memory (random access memory, RAM); the memory may also include non-volatile memory (non-volatile memory), such as read-only memory (read-only memory). , ROM), flash memory (flash memory), hard disk drive (HDD) or solid-state drive (SSD); the memory may also include a combination of the above types of memory.
  • volatile memory volatile memory
  • non-volatile memory non-volatile memory
  • read-only memory read-only memory
  • ROM read-only memory
  • flash memory flash memory
  • HDD hard disk drive
  • SSD solid-state drive
  • Computer-readable instructions are stored in the memory.
  • the computer-readable instructions include multiple software modules, such as a first acquisition module 822, a second acquisition module 824, a third acquisition module 826, a fourth acquisition module 828, and a fifth acquisition module. 830, execute module 832.
  • the processor 810 After the processor 810 executes each software module, it can perform corresponding operations according to the instructions of each software module.
  • an operation performed by a software module actually refers to an operation performed by the processor 810 according to an instruction of the software module.
  • the first acquisition module 822 may be configured to acquire a first flow characteristic, the first flow characteristic includes statistical information generated when the first device forwards traffic according to the first congestion control parameter in the first cycle, and the first congestion control parameter is based on the first congestion control parameter. Obtained by congestion control rules.
  • the second obtaining module 824 is configured to obtain the first reward value according to the first traffic characteristic.
  • the third obtaining module 826 is configured to modify the first congestion control parameter by using the first step to obtain the second congestion control parameter.
  • the fourth obtaining module 828 is configured to obtain the second flow characteristic, and the second flow characteristic includes the statistical information generated when the first device forwards the flow according to the second congestion control parameter in the second period.
  • the fifth obtaining module 830 is configured to obtain the second reward value according to the second traffic characteristic.
  • the execution module 832 is configured to execute corresponding processing if the second reward value is greater than the first reward value.
  • the processor 810 executes the computer-readable instructions in the memory, it can perform all operations that can be performed by the network device or the first device or the second device according to the instructions of the computer-readable instructions. The operations performed in the embodiment corresponding to FIG. 3 and FIG. 4.
  • the disclosed system, device, and method can be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division, and there may be other divisions in actual implementation, for example, multiple units or components may be combined or It can be integrated into another system, or some features can be ignored or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
  • the functional units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated unit can be implemented in the form of hardware or software functional unit.
  • the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium.
  • the technical solution of the present application essentially or the part that contributes to the existing technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , Including several instructions to make a computer device (which may be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Environmental & Geological Engineering (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

本申请实施例公开了一种拥塞控制方法，可以应用于网络控制领域。本申请实施例方法包括：获取第一流量特征，第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，第一拥塞控制参数是根据第一拥塞控制规则获得的；根据第一流量特征，获得第一回报值；利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数；获取第二流量特征，第二流量特征包括第一设备在第二周期内根据第二拥塞控制参数转发流量时生成的统计信息；根据第二流量特征，获得第二回报值；若第二回报值大于第一回报值，则执行相应的处理。本申请实施例可以得到优于第一拥塞控制参数的第二拥塞控制参数，从而能够提高第一拥塞控制规则的场景泛化性。

Description

拥塞控制方法以及相关设备
本申请要求于2019年12月03日提交中国专利局、申请号为201911223339.7、发明名称为“拥塞控制方法以及相关设备”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请实施例涉及网络控制领域，尤其涉及拥塞控制方法以及相关设备。
背景技术
当前传输控制协议(transmission control protocol，TCP)、远程直接内存访问(remote direct memory access，RDMA)等网络技术在网络领域被广泛使用，而这些网络技术对于网络设备的拥塞指标也越来越高，拥塞指标主要包括时延，吞吐等。
为了控制网络设备的拥塞指标，网络设备会设置拥塞控制参数以控制流量。拥塞控制参数可以通过网络设备上安装的人工智能(artificial intelligence，AI)模型获得，AI模型可以利用初始AI模型通过大量的历史流量特征训练得到。利用AI模型得到拥塞控制参数的流程如下：网络设备采集网络设备上的流量特征，例如出口转发速率、队列深度等，网络设备将采集到的流量特征输送到AI模型进行在线推理，AI模型根据流量特征向网络设备的转发芯片输出拥塞控制参数。然后，转发芯片利用获得的拥塞控制参数控制网络设备的流量。
通常，AI模型需要由大量的历史流量特征训练得到，对于历史流量特征的场景不够广泛的情况下，有限的历史流量特征的场景就不能完全覆盖所有的流量场景，会导致AI模型输出的拥塞控制参数不够理想，因此出现AI模型适应性不足甚至不适应的情况，即场景泛化性问题。
发明内容
本申请实施例提供了一种拥塞控制方法以及相关设备，可以提升拥塞控制规则的场景泛化性。
本申请实施例第一方面提供了一种拥塞控制方法。
网络设备上配置有第一拥塞控制规则，网络设备可以通过该第一拥塞控制规则获取第一拥塞控制参数，在第一设备根据第一拥塞控制参数转发流量后，网络设备可以获取第一流量特征，该第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，网络设备可以根据获取到的第一流量特征，获取第一回报值。在网络设备通过第一拥塞控制规则获取第一拥塞控制参数之后，网络设备可以利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数，因为第一步长不等于0，因此第二拥塞控制参数不等于第一拥塞控制参数。在第一设备根据第二拥塞控制参数转发流量后，网络设备可 以获取第二流量特征，该第二流量特征包括第一设备在第二周期内根据第二拥塞控制参数转发流量时生成的统计信息，网络设备可以根据获取到的第二流量特征，获取第二回报值。网络设备可以确定第二回报值是否大于第一回报值，若网络设备确定第二回报值大于第一回报值，则网络设备执行相应的处理。
本申请实施例中，在经过第一拥塞控制规则获得第一拥塞控制参数后，网络设备可以获取第一流量特征，该第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，第一拥塞控制参数是根据第一拥塞控制规则获得的，网络设备根据第一流量特征可以获得第一回报值，在利用第一步长修改第一拥塞控制参数后，网络设备可以获得第二拥塞控制参数，网络设备还可以获取第二流量特征，该第二流量特征包括第一设备在第二周期内根据第二拥塞控制参数转发流量时生成的统计信息，网络设备根据该第二流量特征，可以获得第二回报值，若第二回报值大于第一回报值，则网络设备执行相应的处理，其中，网络设备利用第一步长修改第一拥塞控制参数，获得了可以获得更大回报值的第二拥塞控制参数，因为第二回报值大于第一回报值，因此第二拥塞控制参数优于第一拥塞控制参数，因此网络设备对第一拥塞控制规则的推理结果进行了优化，提升了第一拥塞控制规则的场景泛化性。
在一种可能的设计中，在网络设备利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数之前，网络设备可以获取到第一设备的第一运行值和第二运行值，第一运行值是第一设备在目标周期内的运行值，第二运行值是第一设备在目标周期前一周期内的运行值，网络设备还可以获取到目标阈值，目标阈值为事先设定的一个值，网络设备确定第一运行值和第二运行值的差值的绝对值是否小于目标阈值，只有当第一运行值和第二运行值的差值的绝对值小于目标阈值，网络设备才会执行利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数的步骤。其中，目标周期是指第一周期的前一周期。
本申请实施例中，在网络设备利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数之前，网络设备先确定是否需要利用第一步长修改第一拥塞控制参数，网络设备通过确定第一设备转发的流量是否趋于稳定来确定是否需要利用第一步长修改第一拥塞控制参数，网络设备确定第一设备转发的流量是否趋于稳定的方法是：网络设备获取第一设备不同周期的第一运行值和第二运行值，网络设备确定第一运行值和第二运行值的差值的绝对值是否小于目标阈值，当第一运行值和第二运行值的差值的绝对值小于目标阈值时，网络设备确定第一设备转发的流量趋于稳定，网络设备从而确定需要利用第一步长修改第一拥塞控制参数，其中，网络设备通过确定第一设备转发的流量是否趋于稳定来确定是否需要利用第一步长修改第一拥塞控制参数，因为当第一设备转发的流量趋于稳定时，网络设备后续修改第一拥塞控制参数的结果才更加准确，即网络设备可以确定第二回报值大于第一回报值的结果是由于网络设备利用第一步长修改第一拥塞控制参数获得的，反过来说，当第一设备转发的流量波动较大时，即使实际上网络设备利用第一步长修改第一拥塞控制参数是有优势的，既第二拥塞控制参数优于第一拥塞控制参数，但是由于第一设备转发的流量波动较大，有可能导致第二回报值小于第一回报值，导致网络设备做出错误的判断，因此网络设备通过确定第一设备转发的流量是否趋于稳定来确定是否需要利用第一步长修 改第一拥塞控制参数的方式，可以提升方案的准确性。
在一种可能的设计中，网络设备可以仅通过第一拥塞控制规则调整初始拥塞控制参数，仅通过第一拥塞控制规则调整初始拥塞控制规则是指，网络设备不利用步长调整网络设备根据第一拥塞控制规则得出的拥塞控制参数，网络设备直接将第一拥塞控制规则得出的拥塞控制参数作为第一设备控制流量转发的参数；在网络设备利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数之前，网络设备可以统计网络设备仅通过第一拥塞控制规则调整初始拥塞控制参数的连续调整次数，当该连续调整次数大于设定阈值N时，网络设备才会利用第一步长修改第一拥塞控制参数，相应地，第一拥塞控制参数为将初始拥塞控制参数根据所述第一拥塞控制规则连续调整N+1次得到的。
本申请实施例中，在网络设备利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数之前，网络设备可以统计网络设备仅通过第一拥塞控制规则调整初始拥塞控制参数的连续调整次数，当该连续调整次数大于设定阈值N时，网络设备可以利用第一步长修改第一拥塞控制参数，其中，在每次网络设备根据第一拥塞控制规则得到拥塞控制参数时，网络设备可以不必每次都利用步长修改拥塞控制参数，因此网络设备可以不利用步长调整网络设备根据第一拥塞控制规则得出的拥塞控制参数，而是直接将第一拥塞控制规则得出的拥塞控制参数作为第一设备控制流量转发的参数，但是网络设备的判断机制可能不够完善，网络设备的判断机制是指：网络设备选择利用步长修改后的拥塞控制参数作为第一设备控制流量转发的参数还是选择直接将第一拥塞控制规则得出的拥塞控制参数作为第一设备控制流量转发的参数的选择条件；如果网络设备的判断机制不够完善，可能会导致网络设备持续选择直接将第一拥塞控制规则得出的拥塞控制参数作为第一设备控制流量转发的参数，而不选择利用步长修改后的拥塞控制参数作为第一设备控制流量转发的参数，即不执行网络设备利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数的步骤，因为在网络设备确定连续调整次数大于设定阈值N时，网络设备可以利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数，因此提高了网络设备利用步长修改后的拥塞控制参数作为第一设备控制流量转发的参数的概率，因此提高了网络设备对第一拥塞控制参数进行优化的概率，因此提升了第一拥塞控制规则的泛化性。
在一种可能的设计中，第一拥塞控制规则为第一公式，第一公式为Q＝F(A，B)，其中，Q为拥塞控制参数，A为速率，B为设定时延，F(A，B)为跟A和B相关的函数。
本申请实施例中，具体限定了第一拥塞控制规则为第一公式，且具体限定了公式中的部分参数，其中，A为速率，B为设定时延，B为事先设定的值，A为网络设备获取的速率，网络设备可以利用获取的速率和设定时延，通过第一公式获取拥塞控制参数，当获取的速率发生变化时，拥塞控制参数也可能因此而发生变化，因此实现了网络设备根据不同速率得出不同拥塞控制参数，即实现了第一设备对转发流量的动态控制，因此提升了方案的可实现性。
在一种可能的设计中，第一公式为Q＝A×B。
本申请实施例中，具体限定了A和B的关系，提升了方案的可实现性。
在一种可能的设计中，在第二回报值大于第一回报值时，网络设备利用第一步长修改 第一公式，获得第二公式。
本申请实施例中，网络设备在利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数后，网络设备确定第二回报值大于第一回报值，即第二拥塞控制参数优于第一拥塞控制参数，因此，网络设备利用第一步长修改第一拥塞控制规则，其中，网络设备在确定第二拥塞控制参数优于第一拥塞控制参数后，网络设备利用第一步长修改第一公式，网络设备不仅优化了根据第一拥塞控制规则得出的第一拥塞控制参数，还优化了第一拥塞控制规则，因此从根源上提升了第一拥塞控制规则的泛化性。
在一种可能的设计中，网络设备根据获取的第二流量特征，利用第二公式获得第三拥塞控制参数，该第三拥塞控制参数用于第一设备控制转发的流量。
本申请实施例中，网络设备利用第一步长修改第二公式后，网络设备可以根据第二流量特征，利用第二公式获得第三拥塞控制参数，该第三拥塞控制参数可以用于控制第一设备转发的流量，其中，网络设备利用第二公式获得了第三拥塞控制参数，即网络设备利用第二公式替换了第一公式，而第二公式是由第一公式优化而来，因此提高了第一设备的拥塞指标，提升了第一公式的泛化性。
在一种可能的设计中，第一步长为百分比值。
本申请实施例中，第一步长为百分比值，因为在不同的流量模型中，第一拥塞控制参数的数值可能变化比较大，如果第一步长为具体的值，例如20，当第一拥塞控制参数为1000时，第一步长对第一拥塞控制参数的修改会显得太小；当第一拥塞控制参数的范围为0-30，第一拥塞控制参数为10，第一步长为20时，第一步长对第一拥塞控制参数的修改会显的太大，因此第一步长可以采用百分比值，去掉因为第一拥塞控制参数本身大小变化的干扰，因此在实际应用中，第一步长可以应用于不同的第一拥塞控制参数，因此提升了方案在不同场景下的适应性，即提升了方案的泛化性。
在一种可能的设计中，第二公式为Q＝A×B×(1+C)，其中，所述C为第一步长。
本申请实施例中，具体限定了第二公式为Q＝A×B×(1+C)，C为第一步长，因此提升了方案的可实现性。
在一种可能的设计中，在网络设备确认第二回报值大于第一回报值后，网络设备利用第二步长修改第二拥塞控制参数，获得第三拥塞控制参数，在第一设备根据第三拥塞控制参数转发流量后，网络设备可以获取第三流量特征，该第三流量特征包括第一设备在第三周期内根据第三拥塞控制参数转发流量时生成的统计信息，网络设备可以根据获取到的第三流量特征，获取第三回报值。
本申请实施例中，在网络设备确认第二回报值大于第一回报值后，即网络设备确认第二拥塞控制参数优于第一拥塞控制参数后，网络设备利用第二步长正向修改第二拥塞控制参数，获得第三拥塞控制参数，其中，第二拥塞控制参数优于第一拥塞控制参数，网络设备在得到优于第一拥塞控制参数的第二拥塞控制参数后，因为第二拥塞控制参数有可能并不是最优的拥塞控制参数，网络设备在第二拥塞控制参数的基础上做进一步的优化尝试，因此提高了网络设备获取更优的拥塞控制参数的概率，提高了第一拥塞控制规则的泛化性程度。
在一种可能的设计中，若第三回报值小于第二回报值，网络设备可以利用第三步长反向修改第二拥塞控制参数，获得第四拥塞控制参数，在第一设备根据第四拥塞控制参数转发流量后，网络设备可以获取第四流量特征，该第四流量特征包括第一设备在第四周期内根据第四拥塞控制参数转发流量时生成的统计信息，网络设备可以根据获取到的第四流量特征，获取第四回报值。
本申请实施例中，第二拥塞控制参数是一个值，网络设备可以修改这个值让第二拥塞控制参数变大或者变小，在网络设备尝试将第二拥塞控制参数变大或者变小，获得第三拥塞控制参数后，若第三回报值小于第二回报值，则网络设备可以确定第二拥塞控制参数优于第三拥塞控制参数，则网络设备可以确定前面尝试将第二拥塞控制参数变大或者变小的行为是错误的，因此网络设备利用第三步长反向修改第二拥塞控制参数，获得第四拥塞控制参数，此处限定的反向修改只是为了与前述的正向修改做区分，不限定反向修改就是网络设备对第二拥塞控制参数执行变小操作，其中，网络设备在一个方向上尝试获得比第二拥塞控制参数更优的第三拥塞控制参数失败后，网络设备向相反的反向尝试获得比第二拥塞控制参数更优的第四拥塞控制参数，因此提高了网络设备获取更优的拥塞控制参数的概率，提高了第一拥塞控制规则的泛化性程度。
在一种可能的设计中，网络设备利用第三步长修改第二拥塞控制参数获得了第四拥塞控制参数，在第一设备根据第四拥塞控制参数转发流量后，网络设备可以获取第四流量特征，该第四流量特征包括第一设备在第四周期内根据第四拥塞控制参数转发流量时生成的统计信息，网络设备可以根据获取到的第四流量特征，获取第四回报值，若该第四回报值大于第二回报值，则网络设备利用第三步长和第一步长修改第一拥塞控制规则，获得第二拥塞控制规则。
本申请实施例中，网络设备在利用第三步长修改第二拥塞控制参数，获得第四拥塞控制参数后，网络设备确定第四回报值大于第二回报值，即第四拥塞控制参数优于第二拥塞控制参数，并且，第二回报值大于第一回报值，即第二拥塞控制参数优于第一拥塞控制参数因此，网络设备利用第三步长和第一步长修改第一拥塞控制规则，其中，网络设备在确定第四拥塞控制参数优于第二拥塞控制参数后，网络设备利用第三步长和第一步长修改第一拥塞控制规则，网络设备不仅两次优化了根据第一拥塞控制规则得出的第一拥塞控制参数，还优化了第一拥塞控制规则，因此从根源上提升了第一拥塞控制规则的泛化性。
在一种可能的设计中，在网络设备获取到优于第二拥塞控制参数的第四拥塞控制参数，并利用第三步长和第一步长修改第一拥塞控制规则，获得第二拥塞控制规则后，网络设备根据第四流量特征，利用第二拥塞控制规则生成第一设备的新的拥塞控制参数，该新的拥塞控制参数用于第一设备控制转发的流量。
本申请实施例中，网络设备利用第三步长和第一步长修改第二公式后，网络设备可以根据第四流量特征，利用第二拥塞控制规则获得新的拥塞控制参数，该新的拥塞控制参数可以用于控制第一设备转发的流量，其中，网络设备利用新的拥塞控制参数获得了新的拥塞控制参数，即网络设备利用第二拥塞控制规则替换了第一拥塞控制规则，而第二拥塞控制规则是由第一拥塞控制规则优化而来，因此提高了第一设备的拥塞指标，因此提升了第 一拥塞控制规则的泛化性。
在一种可能的设计中，第二步长大于第一步长。
本申请实施例中，第二步长大于第一步长，在网络设备利用第一步长修改第一拥塞控制参数后，网络设备获得了优于第一拥塞控制参数的第二拥塞控制参数，由此证明网络设备对第一拥塞控制参数进行修改的必要性和正确性，因此网络设备可以在利用步长修改第二拥塞控制参数时，利用大于第一步长的第二步长，其中，因为第二步长大于第一步长，因此在拥塞控制参数的取值范围内，网络设备可以更快的确定出网络设备认为的最优的拥塞控制参数，因此可以减少收敛时间。
在一种可能的设计中，网络设备上配置有第三拥塞控制规则，网络设备可以通过该第三拥塞控制规则获取第五拥塞控制参数，在第二设备根据第五拥塞控制参数转发流量后，网络设备可以获取第五流量特征，该第五流量特征包括第二设备在第一周期内根据第五拥塞控制参数转发流量时生成的统计信息，网络设备可以根据获取到的第五流量特征，获取第五回报值；在网络设备通过第三拥塞控制规则获取第五拥塞控制参数之后，网络设备可以利用第四步长修改第五拥塞控制参数，获得第六拥塞控制参数，因为第一步长不等于0，因此第六拥塞控制参数不等于第五拥塞控制参数，在第二设备根据第六拥塞控制参数转发流量后，网络设备可以获取第六流量特征，该第六流量特征包括第二设备在第二周期内根据第六拥塞控制参数转发流量时生成的统计信息，网络设备可以根据获取到的第六流量特征，获取第六回报值；网络设备可以确定第六回报值与第二回报之和是否大于第五回报值与第一回报值之和，若网络设备确定第六回报值与第二回报之和大于第五回报值与第一回报值之和，则网络设备执行相应的所述相应的处理。
本申请实施例中，在经过第三拥塞控制规则获得第五拥塞控制参数后，网络设备可以获取第五流量特征，该第五流量特征包括第二设备在第一周期内根据第五拥塞控制参数转发流量时生成的统计信息，第五拥塞控制参数是根据第三拥塞控制规则获得的，网络设备根据第五流量特征可以获得第五回报值，在利用第四步长修改第三拥塞控制参数后，网络设备可以获得第六拥塞控制参数，网络设备还可以获取第六流量特征，该第六流量特征包括第二设备在第二周期内根据第六拥塞控制参数转发流量时生成的统计信息，网络设备根据该第六流量特征，可以获得第六回报值；网络设备在获取到第一回报值，第二回报值，第五回报值和第六回报值后，网络设备可以确定第六回报值与第二回报之和是否大于第五回报值与第一回报值之和，若网络设备确定第六回报值与第二回报之和大于第五回报值与第一回报值之和，则网络设备执行所述相应的处理，其中，在网络的流量控制中，第一设备和第二设备间的流量可能会相互影响，因此网络设备综合考虑了两台设备上的回报值之和，网络设备利用第一步长修改第一拥塞控制参数，获得了第二拥塞控制参数，网络设备利用第四步长修改第五拥塞控制参数，获得了第六拥塞控制参数，因为第六回报值与第二回报之和大于第五回报值与第一回报值之和，因此网络设备此次修改第一拥塞控制参数和第二拥塞控制参数的收益是正的，因此提升了在多台设备转发的流量控制时，第一拥塞控制规则和第二拥塞控规则的场景泛化性。
在一种可能的设计中，在网络设备确定第六回报值与第二回报之和大于第五回报值与 第一回报值之和之后，网络设备利用第四步长修改第三拥塞控制规则，网络设备利用第一步长修改所述第一拥塞控制规则。
本申请实施例中，网络设备在利用第四步长修改第五拥塞控制参数，获得第五拥塞控制参数后，网络设备确定第六回报值与第二回报之和大于第五回报值与第一回报值之和，即第六拥塞控制参数和第二拥塞控制参数产生的综合回报值优于第五拥塞控制参数和第一拥塞控制参数的综合回报值，因此，网络设备利用第一步长修改第一拥塞控制规则，网络设备利用第四步长修改所述第三拥塞控制规则，其中，网络设备在确定第六回报值与第二回报之和大于第五回报值与第一回报值之和之后，网络设备修改第一拥塞控制规则和第三拥塞控制规则，网络设备不仅优化了根据第一拥塞控制规则得出的第一拥塞控制参数和第三拥塞控制参数得出的第五拥塞控制参数，还优化了第一拥塞控制规则和第三拥塞控制规则，因此从根源上提升了拥塞控制规则的泛化性。
在一种可能的设计中，在网络设备确认第六回报值与第二回报之和大于第五回报值与第一回报值之和之后，网络设备利用第二步长修改第二拥塞控制参数，获得第三拥塞控制参数，网络设备利用第五步长修改第六拥塞控制参数，获得第七拥塞控制参数，在第一设备根据第三拥塞控制参数转发流量，第二设备根据第七拥塞控制参数转发流量后，网络设备可以获取第三流量特征和第七流量特征，该第三流量特征包括第一设备在第三周期内根据第三拥塞控制参数转发流量时生成的统计信息，该第七流量特征包括第二设备在第三周期内根据第三拥塞控制参数转发流量时生成的统计信息，网络设备根据第三流量特征，获得第三回报值，根据第七流量特征，获得第七回报值。
本申请实施例中，在网络设备第六回报值与第二回报之和大于第五回报值与第一回报值之和之后，网络设备利用第二步长修改第二拥塞控制参数，获得第三拥塞控制参数，网络设备利用第五步长修改第六拥塞控制参数，获得第七拥塞控制参数，其中，第六拥塞控制参数和第二拥塞控制参数的综合收益大于第五拥塞控制参数和第一拥塞控制参数的综合收益，既第六拥塞控制参数和第二拥塞控制参数的组合产生的回报值大于第五拥塞控制参数和第一拥塞控制参数的的组合产生的回报值，网络设备在得到优于第一拥塞控制参数和第五拥塞控制参数的组合后，但是因为第二拥塞控制参数和第六拥塞控制参数的组合有可能并不是最优的拥塞控制参数组合，网络设备在第二拥塞控制参数和第六拥塞控制参数的组合基础上做进一步的优化尝试，因此提高了网络设备获取优于第二拥塞控制参数和第六拥塞控制参数组合的概率，提高了第一拥塞控制规则和第三拥塞控制规则的泛化性程度。
本申请实施例第二方面提供了一种拥塞控制装置，包括多个功能模块，该多个功能模块可以全部是软件模块或硬件模块，还可以是软件模块和硬件模块的组合，该多个功能模块可以根据实现做不同划分，以能实现上述第一方面及其各实施方式中的方法为准则。
本申请实施例第三方面提供了一种拥塞控制设备，包括存储器和处理器。
其中，所述存储器用于存储程序；
所述处理器用于执行所述存储器中的程序，使得该拥塞控制设备执行如上述第一方面及其任意一项实施方式所述的方法。
本申请实施例第四方面提供了一种计算机存储介质，所述计算机存储介质中存储有指 令，所述指令在计算机上执行时，使得所述计算机执行如上述第一方面及其任意一项实施方式所述的方法。
本申请实施例第五方面提供了一种计算机程序产品，所述计算机程序产品在计算机上执行时，使得所述计算机执行如上述第一方面及其任意一项实施方式所述的方法。
附图说明
图1为本申请实施例的网络框架示意图；
图2为本申请实施例中拥塞控制方法的一个流程示意图；
图3为本申请实施例中拥塞控制方法的另一个流程示意图；
图4为本申请实施例中拥塞控制方法的另一个流程示意图；
图5为本申请实施例中拥塞控制方法的另一个流程示意图；
图6为本申请实施例中拥塞控制装置的一个结构示意图；
图7为本申请实施例中拥塞控制装置的另一个结构示意图；
图8为本申请实施例中拥塞控制设备的一个结构示意图。
具体实施方式
本申请实施例提供了一种拥塞控制方法及相关设备，用于网络控制领域，可以提升拥塞控制规则的场景泛化性。
为了更好的理解本申请实施例中的拥塞控制方法，下面对本申请实施例应用的术语进行描述。
AI模型：AI模型是一种算法，是指经过历史数据训练后，可以通过输入数据得到解的模型，AI模型既可以是主流的深度神经网络模型，也可以是传统机器学习模型。
流量控制：在网络中，当有流量经过某个网络节点时，网络节点便会产生流量，为了保持流量的稳定，即为了流量满足一定的条件，会对网络节点的流量进行控制。网络节点可以是网络设备，例如服务器，交换机，路由器等。
当前TCP、RDMA等网络技术在广域网、数据中心等领域被广泛使用，而这些网络技术对于网络的需求也越来越高，其中吞吐和时延是主要的拥塞指标。例如，TCP网络中，网络设备根据丢弃概率参数来丢包以进行流量控制，RDMA网络中通过动态调节显性拥塞通知(explicit congestion notification，ECN)水线来满足吞吐和时延。例如，ECN机制广泛应用于基于增强RoCE(RDMA over Converged Ethernet)的高性能数据中心网络中，合理的ECN配置对网络流量控制起着关键性作用。ECN水线过高，容易导致交换机队列堆积，网络延迟增加，业务时延变差；ECN水线过低，容易导致网络欠吞吐，业务吞吐量将会降低。合理调节ECN配置对网络提高带宽利用利用率和降低网络延迟有着重要作用。交换机对于队列有三个ECN值可以配置，分别是上水线、下水线和最大标记概率。队列长度的两个门限值定义了标记概率，两个门限值分别为下水线和上水线。当队列长度低于门限值下水线时，ECN不会被标记，等同于实际标记概率为0；当队列长度超过门限值上水线时，所有从该队列传输的网络包都会被进行ECN标记，等同于实际标记概率为1，当队列长度 处于两个门限值之间时，数据包会以与队列长度线性增长的概率被进行ECN标记。以交换机为例说明ECN标记的含义，交换机配置ECN上水线和下水线，当交换机端口出现拥塞时，则会根据ECN阈值确定是否为报文打上ECN标记，接收端则会根据携带ECN标记的报文生成拥塞通知(congestion notification packets，CNP)报文告知源端，源端网卡根据CNP报文数量降低发生速率，从而避免拥塞。
业务和网络流量的多样性，会产生不同的网络流量模型。而在不同网络流量模型下，网络设备如何根据流量的变化动态调节和精确适配拥塞控制参数，从而保障网络性能，成为网络中一个重要挑战目标。
下面将结合附图，对本申请实施例中的技术方案进行描述。示例性的，本申请实施例所涉及附图中的以虚线标识的特征或内容可理解为实施例可选地操作或者可选地结构。
请参阅图1，本申请实施例的网络框架包括：网络设备101，第一设备102，第二设备103；其中，网络设备101分别与第一设备102，第二设备103相连。
网络设备101与第一设备102连接的方式和网络设备101与第二设备103连接的方式既可以是通过有线网络连接，也可以是通过无线网络连接。
在实际应用中，网络设备101可以与更多的设备相连。
网络设备101主要的功能是获取第一设备102或第一设备102与第二设备103的流量特征，流量特征包括第一设备102，或第一设备102与第二设备103根据拥塞控制参数转发流量时生成的统计信息，网络设备101根据获取到的流量特征，利用拥塞控制规则生成拥塞控制参数；网络设备101还可以根据获取到的流量特征，计算回报值，以评价第一设备102或第一设备102与第二设备103的拥塞状况。
第一设备102主要的功能是利用获取的拥塞控制参数控制流量的转发，并根据流量的转发生成流量特征，第二设备103的功能与第一设备102的功能类似。
本申请实施例可以没有网络设备101，当没有网络设备101时，第一设备102可以完成网络设备101的所有功能。
本申请实施例可以没有网络设备101，也没有第二设备103，第一设备102无需与第二设备103联合控制流量的转发，并且第一设备102完成网络设备101的所有功能。
上面对本申请实施例的网络框架进行了描述，下面对本申请实施例的拥塞控制方法进行描述。
请参阅图2，为本申请实施例中的拥塞控制方法的流程示意图。
在步骤201中，网络设备获取第一流量特征,根据第一流量特征数据，获得第一回报值。
网络设备配置有第一拥塞控制规则，当网络设备需要对第一设备的流量进行控制时，网络设备可以根据该第一拥塞控制规则获取第一拥塞控制参数，网络设备通过第一拥塞控制规则获得第一拥塞控制参数后，网络设备可以获取第一流量特征，第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，网络设备可以根据第一流量特征，获得第一回报值。
在步骤202中，网络设备利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数。
网络设备利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数，第二拥塞控制参数不等于第一拥塞控制参数。
在步骤203中，网络设备获取第二流量特征,根据第二流量特征，获得第二回报值。
网络设备利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数后，网络设备可以获取第二流量特征，第二流量特征包括第一设备在第二周期内根据第二拥塞控制参数转发流量时生成的统计信息，网络设备可以根据第二流量特征，获得第二回报值。
在步骤204中，若第二回报值大于第一回报值，则网络设备执行相应的处理。
网络设备在获取到第二回报值和第一回报值后，网络设备可以确定第二回报值是否大于第一回报值，若网络设备确定第二回报值大于第一回报值，则网络设备执行相应的处理。
本申请实施例中，若网络设备确定第二回报值大于第一回报值时，网络设备可以执行多种不同的处理，本申请将这种条件下的处理统称为相应的处理。网络设备执行相应的处理可以认为是接受对第一拥塞控制参数的修改的对应操作。
本申请实施例中，在经过第一拥塞控制规则获得第一拥塞控制参数后，网络设备可以获取第一流量特征，该第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，第一拥塞控制参数是根据第一拥塞控制规则获得的，网络设备根据第一流量特征可以获得第一回报值，在利用第一步长修改第一拥塞控制参数后，网络设备可以获得第二拥塞控制参数，网络设备还可以获取第二流量特征，该第二流量特征包括第一设备在第二周期内根据第二拥塞控制参数转发流量时生成的统计信息，网络设备根据该第二流量特征，可以获得第二回报值，若第二回报值大于第一回报值，则网络设备执行相应的处理，其中，网络设备利用第一步长修改第一拥塞控制参数，获得了可以获得更大回报值的第二拥塞控制参数，因为第二回报值大于第一回报值，因此第二拥塞控制参数优于第一拥塞控制参数，因此网络设备对第一拥塞控制规则的推理结果进行了优化，因此提升了第一拥塞控制规则的场景泛化性。
在本申请实施例的拥塞控制方法中，第一拥塞控制参数可以是一个参数，第一拥塞控制参数也可以是多个参数，下面分别对其描述。
一、第一拥塞控制参数是一个参数。
请参阅图3，为本申请提供的拥塞控制方法的另一个实施例的流程示意图。
在步骤301中，网络设备获取第一拥塞控制参数。
网络设备配置有第一拥塞控制规则，第一拥塞控制规则是第一公式，当网络设备需要对第一设备的流量进行控制时，网络设备可以通过向第一公式输入采集的流量特征，获得第一公式输出的第一拥塞控制参数，为了便于理解本实施例中的拥塞控制方法，本实施例以以下公式作为第一公式：
Q＝F(A，B)；
其中，Q为拥塞控制参数，A为速率，B为设定时延。
速率A可以为第一设备或第一设备的某个队列的转发速率，转发速率属于流量特征的一种，流量特征是指第一设备转发流量时生成的统计信息，例如出报文字节数，出报文数，队列深度，被ECN标记的报文个数，吞吐信息，丢包个数等。
设定时延B是事先设定的一个数值，第一设备的拥塞控制配置人员可以根据经验或者模板对设定时延的数值进行设置，设定时延的数值大小体现了配置人员对第一设备的拥塞指标中时延方面的要求。
第一公式具体可以为Q＝A×B。
具体获取第一拥塞控制参数的流程可以如下：网络设备采集第一设备上一周期的速率A，获取事先设定的设定时延B，然后将速率A和设定时延B套入第一公式，获得第一拥塞控制参数Q。
可选地，当没有上一周期的速率A时，网络设备可以采用事先设定的速率，或者在本周期不获取第一拥塞控制参数，等待下一周期获取本周期的速率，利用本周期的速率获得第一拥塞控制参数对第一设备下一周期的流量进行控制。
可选地，第一拥塞控制参数可以是标记概率、丢弃概率。
第一拥塞控制规则可以是AI模型，在实际应用中，可以选择一种AI模型作为第一拥塞控制规则，为了便于说明，本实施例以第一拥塞控制规则是第一公式为例进行说明。
可选地，网络设备可以是第一设备，也可以是除第一设备以外的设备。
在步骤302中，网络设备获取第一流量特征，第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，网络设备根据第一流量特征，获得第一回报值。
网络设备通过第一拥塞控制规则获得第一拥塞控制参数后，网络设备可以获取到第一流量特征，第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，第一流量特征具体可以是第一设备的统计信息，或第一设备的端口的统计信息，或第一设备的队列的统计信息；网络设备可以根据第一流量特征，获得第一回报值。
为了便于理解说明，本实施例的描述引入周期的概念，第一设备利用第一拥塞控制参数在第一周期内对第一设备转发的流量进行控制，第一设备利用第二拥塞控制参数在第二周期内对第一设备转发的流量进行控制，第一流量特征包括第一设备在第一周期内利用第一拥塞控制参数转发流量时生成的统计信息，第二流量特征包括第一设备在第一周期内利用第一拥塞控制参数转发流量时生成的统计信息，依次类推。
第一周期和第二周期的周期时长可以相同，也可以不同，第一周期与第二周期之间可以有间隔，可以相邻，也可以有重叠。
第一设备利用第一拥塞控制参数对第一设备第一周期内转发的流量进行控制后，网络设备可以获取到第一流量特征，第一流量特征可以是在第一周期内某个瞬间的数值，也可以经过处理后的数值，例如，第一流量特征可以是在第一周期内的平均值。
可选地，当网络设备是第一设备时，网络设备直接向转发芯片下发第一拥塞控制参数，让转发芯片利用第一拥塞控制参数控制第一设备在第一周期内转发的流量。
可选地，当网络设备是除第一设备以外的设备时，网络设备在获取到第一拥塞控制参数后，网络设备向第一设备发送第一拥塞控制参数，让第一设备在第一周期内利用第一拥塞控制参数控制第一设备在第一周期内转发的流量。
拥塞指标会包含多方面的指标，例如时延和吞吐等，为了便于评判第一设备不同周期 的拥塞指标的优劣，将多方面的指标整合成一个方面的指标，这一方面的指标便是回报值。
可选地，第一流量特征经过预处理获得吞吐或者时延，例如当第一流量特征为队列深度，可以经过预处理将队列深度转化为时延，例如当第一流量特征为出报文字节数，可以经过预处理将出报文字节数转化为吞吐，然后根据预处理后获得的吞吐或者时延计算第一回报值。
可选地，当拥塞指标只包含时延和吞吐两方面的指标时，网络设备可以采用以下算法获得第一回报值：
R＝m×J+n×K；
其中，R为第一回报值，J为吞吐，K为时延，m和n为加权系数。
可选地，当拥塞指标包含三个方面的指标时，网络设备可以采用以下算法获得第一回报值：
R＝m×J+n×K+v×L；
其中，R为第一回报值，J为吞吐，K为时延，m和n为加权系数，L为业务性能指标，v为加权系数。
在303中、网络设备获取所述第一设备的第一运行值和第二运行值，确定第一运行值和第二运行值的差值是否小于目标阈值。
目标阈值是事先设定的一个值，当第一运行值和第二运行值的差值小于目标阈值时，网络设备可以利用第一步长修改第一拥塞控制参数。
可选地，第一运行值是第一设备在目标周期内的转发速率，目标周期是指第一周期的前一周期，第二运行值是第一设备在目标周期前一周期内的转发速率；或第一运行值是第一设备的第一队列在目标周期内的转发速率，第二运行值是第一设备的第一队列在目标周期前一周期内的转发速率。
可选地，第一运行值是第一设备的第一队列在目标周期内的队列深度，目标周期是指第一周期的前一周期，第一运行值是第一设备的第一队列在目标周期前一周期内的队列深度。
当第一运行值不同时，目标阈值也可以不相同，例如当第一运行值是第一设备在目标周期内的转发速率，目标阈值是第一阈值，当第一运行值是第一设备的第一队列在目标周期内的转发速率，目标阈值是第二阈值，第一阈值不等于第二阈值。
可选地，网络设备可以仅通过第一拥塞控制规则调整初始拥塞控制参数，仅通过第一拥塞控制规则调整初始拥塞控制规则是指，网络设备不利用步长调整网络设备根据第一拥塞控制规则得出的拥塞控制参数，网络设备直接将第一拥塞控制规则得出的拥塞控制参数作为第一设备控制流量转发的参数；在网络设备利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数之前，网络设备可以统计网络设备仅通过第一拥塞控制规则调整初始拥塞控制参数的连续调整次数。
在步骤304中、若第一运行值和第二运行值的差值小于目标阈值，则网络设备利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数。
若第一运行值和第二运行值的差值小于目标阈值，则网络设备利用第一步长正向修改 第一拥塞控制参数，获得第二拥塞控制参数。
正向修改只是为了和后面描述的反向修改做区别，并不是指第一拥塞控制参数只能正向加的意思，例如，第一拥塞控制参数为100，第一步长为20，网络设备利用第一步长正向修改第一拥塞控制参数后，第二拥塞控制参数可以是80，也可以是120，为了方便理解本实施例的拥塞控制方法，本实施例以第二拥塞控制参数是120为例进行说明。
可选地，第一步长为百分比值，因为在网络中，第一拥塞控制参数可能变化比较大，如果第一步长为具体的数值，例如20，当第一拥塞控制参数为1000时，第一步长对第一拥塞控制参数的修改会显得太小，因此第一步长可以采用百分比值，去掉因为第一拥塞控制参数变化的干扰，当第一拥塞控制参数为100，第一步长为20％时，网络设备利用第一步长正向修改第一拥塞控制参数，获得的第二步长可以是80，也可以是120。
可选地，目标阈值可以根据第一设备的网络波动大小进行调节，当第一设备的网络波动较小时，即第一设备的流量变化较小，网络设备将目标阈值调小，当第一设备的网络波动较大时，即第一设备的流量变化较大，网络设备将目标阈值调大。
可选地，当网络设备统计网络设备可以仅通过第一拥塞控制规则调整初始拥塞控制参数的连续调整次数时，网络设备可以不获取第一设备的第一运行值和第二运行值，网络设备可以确定该连续调整次数是否大于设定阈值N，当网络设备确定该连续调整次数大于设定阈值N时，则网络设备执行利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数的步骤。
可选地，当网络设备既统计网络设备仅通过第一拥塞控制规则调整初始拥塞控制参数的连续调整次数，网络设备又获取了第一设备的第一运行值和第二运行值时，只要有一个条件满足，则网络设备执行利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数的步骤，即包括以下几种情况：
当网络设备确定第一运行值和第二运行值的差值大于目标阈值，网络设备确定该连续调整次数大于设定阈值N时，网络设备执行利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数的步骤。
当网络设备确定第一运行值和第二运行值的差值小于目标阈值，网络设备确定该连续调整次数小于设定阈值N时，网络设备执行利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数的步骤。
当网络设备确定第一运行值和第二运行值的差值小于目标阈值，网络设备确定该连续调整次数大于设定阈值N时，网络设备执行利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数的步骤。
可选地，步骤303可以不执行，当不执行步骤303时，网络设备获取到第一拥塞控制参数后，网络设备可以无需确认第一运行值和第二运行值的差值小于目标阈值，直接利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数。
在步骤305中，网络设备获取第二流量特征，第二流量特征包括所述第一设备在第二周期内根据第二拥塞控制参数转发流量时生成的统计信息；网络设备根据第二流量特征，获得第二回报值。
网络设备利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数后，网络设备可以获取第二流量特征，第二流量特征包括所述第一设备在第二周期内根据第二拥塞控制参数转发流量时生成的统计信息，第二流量特征具体可以是第一设备的统计信息，或第一设备的端口的统计信息，或第一设备的队列的统计信息；网络设备获取第二流量特征后，网络设备可以根据获取到的第二流量特征获得第二回报值，第二回报值的算法与步骤302中第一回报值的算法类似。第二流量特征可以是第一设备在第二周期内某个瞬间的数值，也可以经过处理后的数值，例如，第二流量特征可以是第一设备在第二期内的平均值。
可选地，当网络设备为第一设备时，网络设备直接向转发芯片下发第二拥塞控制参数，让转发芯片控制第一设备在第二周期转发的流量。
可选地，当网络设备为除第一设备以外的设备时，网络设备在获取到第二拥塞控制参数后，网络设备会向第一设备发送第二拥塞控制参数，用于让第一设备利用第二拥塞控制参数来控制第一设备在第二周期转发的流量。
在步骤306中，网络设备确认第二回报值是否大于第一回报值。
网络设备获取第二回报值和第一回报值后，网络设备确认第二回报值是否大于第一回报值。
在步骤307中，若第二回报值大于第一回报值，则网络设备利用第一步长修改第一公式，获得第二公式，利用第二公式获得拥塞控制参数。
若第二回报值大于第一回报值，则网络设备利用第一步长修改第一公式，获得第二公式，利用第二公式获得拥塞控制参数。
可选地，第二公式可以是以下公式：
Q＝A×B×(1+C1)；
其中，Q为拥塞控制参数，A为速率，B为设定时延，C1为百分比值的第一步长。
网络设备利用第一设备在第一周期的速率，设定时延和第一步长，可以通过第二公式获得拥塞控制参数，将该拥塞控制参数作为第一拥塞控制参数，将第二公式作为第一公式，返回步骤302。
可选地，在网络设备确定第二回报值大于第一回报值时，网络设备先不用第一步长修改第一公式，获得第二公式的步骤，而是用第二步长继续正向修改第二拥塞控制参数，获得第三拥塞控制参数，网络设备获取第三流量特征，第三流量特征包括所述第一设备在第三周期内根据第三拥塞控制参数转发流量时生成的统计信息；网络设备根据第三流量特征，获得第三回报值；若第三回报值大于第二回报值，则网络设备继续用第三步长正向修改第三拥塞控制参数，获得第四拥塞控制参数，网络设备获取第四流量特征，第四流量特征包括所述第一设备在第四周期内根据第四拥塞控制参数转发流量时生成的统计信息；网络设备根据第四流量特征，获得第四回报值，依次类推，直到第T+1回报值小于第T回报值，网络设备用第一步长，第二步长，至第T－1步长中所有步长的和作为第一步长，执行用第一步长修改第一公式，获得第二公式的步骤。
在步骤308中、若第二回报值小于第一回报值，则网络设备利用第二步长反向修改第一拥塞控制参数，获得第三拥塞控制参数。
若第二回报值小于第一回报值，则网络设备利用第二步长反向修改第一拥塞控制参数，获得第三拥塞控制参数。
反向修改只是为了和前面描述的正向修改做区别，并不是指第一拥塞控制参数只能反向减的意思，例如，第一拥塞控制参数为100，第二步长为20，网络设备利用第二步长反向修改第一拥塞控制参数后，第三拥塞控制参数可以是80，也可以是120，为了方便理解本实施例的拥塞控制方法，本实施例以第三拥塞控制参数是80为例进行说明。
可选地，第二步长可以等于第一步长，例如，第一步长为20％，第二步长也可以是20％。
在步骤309中、网络设备获取第三流量特征，第三流量特征包括所述第一设备在第三周期内根据第三拥塞控制参数转发流量时生成的统计信息；网络设备根据第三流量特征，获得第三回报值。
网络设备利用第一步长反向修改第一拥塞控制参数，获得第三拥塞控制参数后，网络设备获取第三流量特征，第三流量特征包括所述第一设备在第三周期内根据第三拥塞控制参数转发流量时生成的统计信息，第三流量特征具体可以是第一设备的统计信息，或第一设备的端口的统计信息，或第一设备的队列的统计信息；网络设备获取第三流量特征后，网络设备可以根据获取到的第三流量特征获取第三回报值，第三回报值的算法与步骤302中第一回报值的算法类似。
在步骤310中，网络设备确认第三回报值是否大于第一回报值。
网络设备获取第三回报值和第一回报值后，网络设备确认第三回报值是否大于第一回报值。
在步骤311中，若第三回报值大于第一回报值，则网络设备利用第二步长修改第一公式，获得第二公式，利用第二公式获得新的拥塞控制参数。
若第三回报值大于第一回报值，则网络设备利用第二步长修改第一公式，获得第二公式，利用第二公式获得拥塞控制参数。
可选地，第二公式可以是以下算法：
Q＝A×B×(1+C2)；
其中，Q为拥塞控制参数，A为速率，B为设定时延，C2为百分比值的第二步长。
网络设备利用第一设备在第二周期的速率，设定时延和第一步长，可以通过第二公式获得拥塞控制参数，将该拥塞控制参数作为第一拥塞控制参数，将第二公式作为第一公式，返回步骤302。
可选地，在网络设备确定第三回报值大于第一回报值时，网络设备先不用第一步长修改第一公式，获得第二公式的步骤，而是用第三步长继续反向修改第三拥塞控制参数，获得第四拥塞控制参数，网络设备获取第四流量特征，第四流量特征包括所述第一设备在第四周期内根据第四拥塞控制参数转发流量时生成的统计信息；网络设备根据第四流量特征，获得第四回报值；若第四回报值大于第三回报值，则网络设备继续用第四步长反向修改第四拥塞控制参数，获得第五拥塞控制参数，网络设备获取第五流量特征，第五流量特征包括所述第一设备在第五周期内根据第五拥塞控制参数转发流量时生成的统计信息；网络设备根据第五流量特征，获得第五回报值，依次类推，直到第T+1回报值小于第T回报值， 网络设备将第一步长，第二步长，至第T－1步长中所有步长的和作为第二步长，执行用第二步长修改第一公式，获得第二公式的步骤。
在步骤312中，若第三回报值小于第一回报值，则网络设备利用第三步长正向修改第一拥塞控制参数进行，获得第四拥塞控制参数。
若第三回报值小于第一回报值，则网络设备利用第三步长正向修改第一拥塞控制参数，获得第四拥塞控制参数。
可选地，第三步长大于第一步长，因为网络设备正向修改和反向修改第一拥塞控制参数后，获得的第二回报值和第三回报值都小于第一回报值，那么有概率陷入了局部最优，假设第一拥塞控制参数为100，第二拥塞控制参数为120，第二拥塞控制参数为80，有概率在区间80至120之间，100附近为最优的解，为了跳出局部最优，第三步长要大于第一步长。
在步骤313中，网络设备获取第四流量特征，第四流量特征包括所述第一设备在第四周期内根据第四拥塞控制参数转发流量时生成的统计信息；网络设备根据第四流量特征，获得第四回报值。
网络设备利用第三步长正向修改第一拥塞控制参数，获得第四拥塞控制参数后，网络设备获取第四流量特征，第四流量特征包括所述第一设备在第四周期内根据第四拥塞控制参数转发流量时生成的统计信息，第四流量特征具体可以是第一设备的统计信息，或第一设备的端口的统计信息，或第一设备的队列的统计信息；网络设备获取第四流量特征后，网络设备可以根据获取到的第四流量特征获得第四回报值，第四回报值的算法与步骤302中第一回报值的算法类似。
在步骤314中，网络设备确认第四回报值是否大于第一回报值。
网络设备获取第四回报值和第一回报值后，网络设备确认第四回报值是否大于第一回报值。
在步骤315中，若第四回报值大于第一回报值，则网络设备利用第三步长修改第一公式，获得第二公式，利用第二公式获得新的拥塞控制参数。
若第四回报值大于第一回报值，则网络设备利用第三步长修改第一公式，获得第二公式，利用第二公式获得新的拥塞控制参数。
可选地，第二公式可以是以下算法：
Q＝A×B×(1+C3)；
其中，Q为拥塞控制参数，A为速率，B为设定时延，C3为百分比值的第三步长。
网络设备利用第一设备在第四周期的速率，设定时延和第一步长，可以通过第二公式获得拥塞控制参数，将该拥塞控制参数作为第一拥塞控制参数，将第二公式作为第一公式，返回步骤302。
可选地，在网络设备确定第四回报值大于第一回报值时，网络设备先不用第三步长修改第一公式，获得第二公式的步骤，而是用第四步长继续正向修改第四拥塞控制参数，获得第五拥塞控制参数，网络设备获取第五流量特征，第五流量特征包括所述第一设备在第五周期内根据第五拥塞控制参数转发流量时生成的统计信息；网络设备根据第五流量特征， 获得第五回报值；若第五回报值大于第四回报值，则网络设备继续用第五步长正向修改第五拥塞控制参数，获得第六拥塞控制参数，网络设备获取第六流量特征，第六流量特征包括所述第一设备在第六周期内根据第六拥塞控制参数转发流量时生成的统计信息；网络设备根据第六流量特征，获得第六回报值，依次类推，直到第T+1回报值小于第T回报值，网络设备将第一步长，第二步长，至第T－1步长中所有步长的和作为第三步长，执行用第三步长修改第一公式，获得第二公式的步骤。
在步骤316中，若第四回报值小于第一回报值，则网络设备利用第四步长反向修改第一拥塞控制参数，获得第五拥塞控制参数。
若第四回报值小于第一回报值，则网络设备利用第四步长反向修改第一拥塞控制参数，获得第五拥塞控制参数。
可选地，第四步长大于第二步长，因为网络设备正向修改和反向修改第一拥塞控制参数后，获得的第二回报值和第三回报值都小于第一回报值，那么有概率陷入了局部最优，假设第一拥塞控制参数为100，第二拥塞控制参数为120，第二拥塞控制参数为80，有概率在区间80至120之间，100附近为最优的解，为了跳出局部最优，第四步长要大于第二步长。
在步骤317中，网络设备获取第五流量特征，第五流量特征包括所述第一设备在第五周期内根据第五拥塞控制参数转发流量时生成的统计信息；网络设备根据第五流量特征，获得第五回报值。
网络设备利用第四步长反向修改第一拥塞控制参数，获得第五拥塞控制参数后，网络设备获取第五流量特征，第五流量特征包括所述第一设备在第五周期内根据第五拥塞控制参数转发流量时生成的统计信息，第五流量特征具体可以是第一设备的统计信息，或第一设备的端口的统计信息，或第一设备的队列的统计信息；网络设备获取第五流量特征后，网络设备可以根据获取到的第五流量特征获得第五回报值，第五回报值的算法与步骤302中第一回报值的算法类似。
在步骤318中，网络设备确认第五回报值是否大于第一回报值。
网络设备获取第五回报值和第一回报值后，网络设备确认第五回报值是否大于第一回报值。
在步骤319中，若第五回报值大于第一回报值，则网络设备利用第四步长修改第一公式，获得第二公式，利用第二公式获得新的拥塞控制参数。
若第五回报值大于第一回报值，则网络设备利用第四步长修改第一公式，获得第二公式，利用第二公式获得拥塞控制参数。
可选地，第二公式可以是以下算法：
Q＝A×B×(1+C4)；
其中，Q为拥塞控制参数，A为速率，B为设定时延，C4为百分比值的第四步长。
网络设备利用第一设备在第五周期的速率，设定时延和第一步长，可以通过第二公式获得拥塞控制参数，将该拥塞控制参数作为第一拥塞控制参数，将第二公式作为第一公式，返回步骤302。
可选地，在网络设备确定第五回报值大于第一回报值时，网络设备先不用第四步长修改第一公式，获得第二公式的步骤，而是用第五步长继续反向修改第五拥塞控制参数，获得第六拥塞控制参数，网络设备获取第六流量特征，第六流量特征包括所述第一设备在第六周期内根据第六拥塞控制参数转发流量时生成的统计信息；网络设备根据第六流量特征，获得第六回报值；若第六回报值大于第五回报值，则网络设备继续用第六步长反向修改第六拥塞控制参数，获得第七拥塞控制参数，网络设备获取第七流量特征，第七流量特征包括所述第一设备在第七周期内根据第七拥塞控制参数转发流量时生成的统计信息；网络设备根据第七流量特征，获得第七回报值，依次类推，直到第T+1回报值小于第T回报值，网络设备将第一步长，第二步长，至第T－1步长中所有步长的和作为第四步长，执行用第四步长修改第一公式，获得第二公式的步骤。
在步骤320中、若第五回报值大于第一回报值，网络设备有百分之F的概率将第三步长作为第一步长，将第四步长作为第二步长，返回步骤312，网络设备有百分之G的概率返回步骤301。
若第五回报值大于第一回报值，网络设备有百分之F的概率将第三步长作为第一步长，将第四步长作为第二步长，返回步骤312，网络设备有百分之G的概率返回步骤301。
可选地，F加G等于100。
可选地，F的数值随着返回步骤312的次数而衰减，例如第一次F等于50，步骤320后，返回了步骤312，从步骤312又到了步骤320，F的数值变为40，则在此次判定中，有百分之40的概率会返回步骤312。
二、第一拥塞控制参数是多个参数。
为了便于理解说明，下面以动态调节ECN配置为例进行说明，第一拥塞控制参数包括3个参数，分别是下水线、上水线、最大标记概率。
请参阅图4，为本申请提供的拥塞控制方法的另一个流程示意图。
在步骤401中，网络设备获取第一拥塞控制参数。
网络设备配置有第一拥塞控制规则，第一拥塞控制规则是AI模型，当网络设备需要对第一设备的流量进行控制时，网络设备可以通过向AI模型输入采集的初始流量特征，获得AI模型输出的第一拥塞控制参数，第一拥塞控制参数包括第一下水线、第一上水线、第一最大标记概率。
可选地，当没有上一周期的流量特征时，网络设备可以采用事先设定的流量特征作为初始流量特征，或者在本周期不获取第一拥塞控制参数，等待下一周期获取本周期的流量特征作为初始流量特征，利用本周期的流量特征获得第一拥塞控制参数对第一设备下一周期的流量进行控制。
第一拥塞控制规则还可以是公式，在实际应用中，可以选择一种公式作为第一拥塞控制规则，为了便于说明，本实施例以第一拥塞控制规则是AI模型为例进行说明。
可选地，网络设备可以是第一设备，也可以是除第一设备以外的设备。
在步骤402中，网络设备获取第一流量特征，第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，网络设备根据第一流量特征，获得 第一回报值。
在步骤403中，网络设备获取所述第一设备的第一运行值和第二运行值，确定第一运行值和第二运行值的差值是否小于目标阈值。
步骤402、步骤403与前述图3中的步骤302、步骤303类似。
在步骤404中，若第一运行值和第二运行值的差值小于目标阈值，则网络设备利用第一步长正向修改第一下水线，获得第二正向下水线。
若第一运行值和第二运行值的差值小于目标阈值，则网络设备利用第一步长正向修改第一下水线，获得第二正向下水线，网络设备将第二正向下水线，第一上水线以及第一最大标记概率作为第二拥塞控制参数，其它描述与前述图3中的步骤304类似。
在步骤405中，网络设备获取第二流量特征，第二流量特征包括所述第一设备根据第二拥塞控制参数转发流量时生成的统计信息；网络设备根据第二流量特征，获得第二回报值。
步骤405前述图3中的步骤305类似。
在步骤406中，网络设备利用第二步长正向修改第一上水线，获得第二正向上水线。
网络设备利用第二步长正向修改第一上水线，获得第二正向上水线，网络设备将目标下水线，第二正向上水线以及第一最大标记概率作为第三拥塞控制参数，目标下水线包括第一下水线或第二正向下水线，若第二回报值大于第一回报值，则目标下水线包括第二正向下水线，若第二回报值大于第一回报值，则目标下水线包括第一下水线。
可选地，若第二回报值大于第一回报值，网络设备可以利用第二步长正向修改第二正向下水线，获得第三正向下水线，网络设备将第三正向下水线，第一上水线以及第一最大标记概率作为第三拥塞控制参数，网络设备获取第三流量特征，第三流量特征包括所述第一设备在第三周期内根据第三拥塞控制参数转发流量时生成的统计信息；网络设备根据第三流量特征，获得第三回报值；若第三回报值大于第二回报值，则网络设备继续用第三步长正向修改第三正向下水线，获得第四正向下水线，网络设备将第四正向下水线，第一上水线以及第一最大标记概率作为第四拥塞控制参数，网络设备获取第四流量特征，第四流量特征包括所述第一设备在第四周期内根据第四拥塞控制参数转发流量时生成的统计信息；网络设备根据第四流量特征，获得第四回报值，依次类推，直到第T+1回报值小于第T回报值，网络设备将第一步长，第二步长，至第T－1步长中所有步长的和作为下水线正向步长，第T回报值对应第T正向下水线，网络设备将第T正向下水线作为第二正向下水线。
可选地，若第二回报值小于第一回报值，可以得出网络设备利用第一步长正向修改获得的第二正向下水线劣于第一下水线，网络设备利用第二正向下水线修改AI模型，在AI模型输入相同的初始流量特征时，减小AI模型输出第二正向下水线的概率，例如，在初始流量特征为50时，向AI模型输入初始流量特征50，AI模型输出第一下水线10，第一上水线60，第一最大标记概率50的概率为30％，AI模型输出第二正向下水线12，第一上水线30，第一最大标记概率50的概率为7％，网络设备修改AI模型，AI模型在输入初始流量特征50时，AI模型输出第二正向下水线12，第一上水线30，第一最大标记概率50的概率为3％。因此网络设备下次获取到初始流量特征50，利用AI模型推理获得拥塞控制参 数时，AI模型输出第二正向下水线12，第一上水线30，第一最大标记概率50的概率会降低。
网络设备利用第二步长正向修改第一上水线和网络设备利用第二步长正向修改第二正向下水线中都出现了第二步长，但是这第二步长只是了方便描述，并不是限定这两个第二步长的数值要相等，相似的，本实施例中，即使出现了相同的步长，还要确定步长修改的对象是否相同，只有步长相同，且步长修改的对象相同，才可以确定两个步长的数值相同，例如，网络设备利用第二步长正向修改第一上水线的修改的对象是第一上水线，网络设备利用第二步长正向修改第二正向下水线的修改的对象是第二正向下水线，因此这两个步长的数值可以不相同。
在步骤407中，网络设备获取第三流量特征，第三流量特征包括所述第一设备根据第三拥塞控制参数转发流量时生成的统计信息；网络设备根据第三流量特征，获得第三回报值。
网络设备获取第三流量特征，第三流量特征包括所述第一设备根据第三拥塞控制参数转发流量时生成的统计信息；网络设备根据第三流量特征，获得第三回报值。
在步骤408中，网络设备利用第三步长正向修改第一最大标记概率，获得第二正向最大标记概率。
网络设备利用第三步长正向修改第一最大标记概率，获得第二正向最大标记概率；网络设备将目标下水线，目标上水线以及第二最大标记概率作为第四拥塞控制参数，目标下水线包括第一下水线或第二正向下水线，若第二回报值大于第一回报值，则目标下水线包括第二正向下水线，若第二回报值大于第一回报值，则目标下水线包括第一下水线，目标上水线包括第一上水线或第二正向上水线，若第三回报值大于第一回报值，且第三回报值大于第二回报值，则目标上水线包括第二正向上水线，否则目标上水线包括第一上水线。
可选地，与步骤406类似，若第三回报值大于第二回报值，且第三回报值大于第一回报值，网络设备可以利用第三步长正向修改第二正向上水线。
在步骤409中，网络设备获取第四流量特征，第四流量特征包括所述第一设备根据第四拥塞控制参数转发流量时生成的统计信息；网络设备根据第四流量特征，获得第四回报值。
网络设备获取第四流量特征，第四流量特征包括所述第一设备根据第四拥塞控制参数转发流量时生成的统计信息；网络设备根据第四流量特征，获得第四回报值。
在步骤410中，网络设备利用第四步长反向修改第一下水线，获得第二反向下水线。
网络设备利用第四步长反向修改第一下水线，获得第二反向下水线，网络设备将第二反向下水线，目标上水线以及目标最大标记概率作为第五拥塞控制参数，目标上水线包括第一上水线或第二正向上水线，若第三回报值大于第一回报值，且第三回报值大于第二回报值，则目标上水线包括第二正向上水线，否则目标上水线包括第一上水线，目标最大标记概率包括第一最大标记概率或第二正向最大标记概率，若第四回报值大于第三回报值，且第四回报值大于第二回报值，且第四回报值大于第一回报值，则目标最大标记概率包括第二正向最大标记概率，否则目标最大标记概率包括第一最大标记概率。
可选地，与步骤406类似，若第四回报值大于第三回报值，且第四回报值大于第二回报值，且第四回报值大于第一回报值，网络设备可以利用第四步长正向修改第二正向最大标记概率。
在步骤411中，网络设备获取第五流量特征，第五流量特征包括所述第一设备根据第五拥塞控制参数转发流量时生成的统计信息；网络设备根据第五流量特征，获得第五回报值。
网络设备获取第五流量特征，第五流量特征包括所述第一设备根据第五拥塞控制参数转发流量时生成的统计信息；网络设备根据第五流量特征，获得第五回报值。
在步骤412中，网络设备利用第五步长反向修改第一上水线，获得第二反向上水线。
网络设备利用第五步长反向修改第一上水线，获得第二反向上水线，网络设备将反向目标下水线，第二反向上水线以及目标最大标记概率作为第六拥塞控制参数，反向目标下水线包括目标下水线或第二反向下水线，若第五回报值大于第四回报值，且第五回报值大于第三回报值，且第五回报值大于第二回报值，且第五回报值大于第一回报值，则反向目标下水线包括第二反向下水线，否则反向目标下水线包括目标下水线，目标最大标记概率包括第一最大标记概率或第二正向最大标记概率，若第四回报值大于第三回报值，且第四回报值大于第二回报值，且第四回报值大于第一回报值，则目标最大标记概率包括第二正向最大标记概率，否则目标最大标记概率包括第一最大标记概率。
可选地，与步骤406类似，若第五回报值大于第四回报值，且第五回报值大于第三回报值，且第五回报值大于第二回报值，且第五回报值大于第一回报值，网络设备可以利用第五步长反向修改第二反向下水线。
在步骤413中、网络设备获取第六流量特征，第六流量特征包括所述第一设备根据第六拥塞控制参数转发流量时生成的统计信息；网络设备根据第六流量特征，获得第六回报值。
网络设备获取第六流量特征，第六流量特征包括所述第一设备根据第六拥塞控制参数转发流量时生成的统计信息；网络设备根据第六流量特征，获得第六回报值。
在步骤414中，网络设备利用第六步长反向修改第一最大标记概率，获得第二反向最大标记概率。
网络设备利用第六步长反向修改第一最大标记概率，获得第二反向最大标记概率，网络设备将反向目标下水线，反向目标上水线以及第二反向最大标记概率作为第七拥塞控制参数，反向目标下水线包括目标下水线或第二反向下水线，若第五回报值大于第四回报值，且第五回报值大于第三回报值，且第五回报值大于第二回报值，且第五回报值大于第一回报值，则反向目标下水线包括第二反向下水线，否则反向目标下水线包括目标下水线，反向目标上水线包括目标上水线或第二反向上水线，若第六回报值大于第五回报值，且第六回报值大于第四回报值，且第六回报值大于第三回报值，且第六回报值大于第二回报值，且第六回报值大于第一回报值，则反向目标上水线包括第二反向上水线，否则反向目标上水线包括目标上水线。
可选地，与步骤406类似，若第六回报值大于第五回报值，且第六回报值大于第四回 报值，且第六回报值大于第三回报值，且第六回报值大于第二回报值，且第六回报值大于第一回报值，网络设备可以利用第六步长反向修改第二反向上水线。
在步骤415中，网络设备获取第七流量特征，第七流量特征包括所述第一设备根据第七拥塞控制参数转发流量时生成的统计信息；网络设备根据第七流量特征，获得第七回报值。
网络设备获取第七流量特征，第七流量特征包括所述第一设备根据第七拥塞控制参数转发流量时生成的统计信息；网络设备根据第七流量特征，获得第七回报值。
在步骤416中、网络设备获取新的初始流量特征，网络设备将新的初始流量特征输入AI模型进行推理，获得初始拥塞控制参数。
网络设备获取新的初始流量特征，网络设备将新的初始流量特征输入AI模型进行推理，获得初始拥塞控制参数，该初始拥塞控制参数用于让第一设备根据该初始拥塞控制参数控制转发的流量，且网络设备不需要利用步长修改初始拥塞控制参数，直接将初始拥塞控制参数作为第一设备控制流量转发的参数。
可选地，当第七回报值大于第一回报值，第七拥塞控制参数不等于第一拥塞控制参数时，网络设备利用第七拥塞控制参数修改AI模型，网络设备利用修改后的AI模型进行推理，获得初始拥塞控制参数。第七拥塞控制参数包括反向目标下水线，反向目标上水线，反向目标最大标记概率。反向目标下水线包括目标下水线或第二反向下水线，若第五回报值大于第四回报值，且第五回报值大于第三回报值，且第五回报值大于第二回报值，且第五回报值大于第一回报值，则反向目标下水线包括第二反向下水线，否则反向目标下水线包括目标下水线，目标下水线包括第一下水线或第二正向下水线，若第二回报值大于第一回报值，则目标下水线包括第二正向下水线，若第二回报值大于第一回报值，则目标下水线包括第一下水线；反向目标上水线包括目标上水线或第二反向上水线，若第六回报值大于第五回报值，且第六回报值大于第四回报值，且第六回报值大于第三回报值，且第六回报值大于第二回报值，且第六回报值大于第一回报值，则反向目标上水线包括第二反向上水线，否则反向目标上水线包括目标上水线，目标上水线包括第一上水线或第二正向上水线，若第三回报值大于第一回报值，且第三回报值大于第二回报值，则目标上水线包括第二正向上水线，否则目标上水线包括第一上水线；反向目标最大标记概率包括目标最大标记概率或第二反向最大标记概率；若第七回报值大于第六回报值，且第七回报值大于第五回报值，且第七回报值大于第四回报值，且第七回报值大于第三回报值，且第七回报值大于第二回报值，且第七回报值大于第一回报值，则反向目标最大标记概率包括第二反向最大标记概率，否则反向目标最大标记概率包括目标最大标记概率，目标最大标记概率包括第一最大标记概率或第二正向最大标记概率，若第四回报值大于第三回报值，且第四回报值大于第二回报值，且第四回报值大于第一回报值，则目标最大标记概率包括第二正向最大标记概率，否则目标最大标记概率包括第一最大标记概率。
例如，在AI模型输入相同的初始流量特征时，增加AI模型输出第七拥塞控制参数的概率，例如，在初始流量特征为50时，向AI模型输入初始流量特征50，AI模型输出第一下水线10，第一上水线60，第一最大标记概率50的概率为30％，AI模型输出第七拥塞控 制参数：第二正向下水线12，第二反向下水线27，第二正向最大标记概率60的概率为7％，网络设备修改AI模型，AI模型在输入初始流量特征50时，AI模型输出第七拥塞控制参数：第二正向下水线12，第二反向下水线27，第二正向最大标记概率60的概率为50％。因此网络设备下次获取到初始流量特征50，利用AI模型推理获得拥塞控制参数时，AI模型输出第七拥塞控制参数：第二正向下水线12，第二反向下水线27，第二正向最大标记概率60的概率会提高，从7％提高到了50％。
可选地，与步骤406类似，若第七回报值大于第六回报值，且第七回报值大于第五回报值，且第七回报值大于第四回报值，且第七回报值大于第三回报值，且第七回报值大于第二回报值，且第七回报值大于第一回报值，网络设备可以利用第七步长反向修改第二反向最大标记概率。
在本申请实施例的拥塞控制方法中，除网络设备以外，可以是单个设备利用拥塞控制方法对设备转发的流量进行控制，也可以是多个设备联合使用拥塞控制方法对设备转发的流量进行控制，上面对单个设备的情况进行了描述，下面多个设备的情况进行描述。
请参阅图5，为本申请提供的拥塞控制方法的另一个实施例的流程示意图。
在步骤501中，网络设备获取第一拥塞控制参数和第二拥塞控制参数。
网络设备配置有第一拥塞控制规则，当网络设备需要对第一设备的流量进行控制时，网络设备可以根据从第一设备采集的第一初始流量特征，利用第一拥塞控制规则获得第一拥塞控制参数。网络设备还配置有第二拥塞控制规则，当网络设备需要对第二设备的流量进行控制时，网络设备可以根据从第二设备采集的第二初始流量特征，利用第二拥塞控制规则获得第二拥塞控制参数。
可选地，当没有上一周期的流量特征时，网络设备可以采用事先设定的流量特征作为初始流量特征，或者在本周期不获取第一拥塞控制参数和第二拥塞控制参数，等待下一周期获取本周期的流量特征作为初始流量特征，利用第一设备本周期的流量特征获得第一拥塞控制参数对第一设备下一周期的流量进行控制，利用第二设备本周期的流量特征获得第二拥塞控制参数对第二设备下一周期的流量进行控制。
第一拥塞控制规则可以是AI模型，第一拥塞控制规则还可以是公式，在实际应用中，可以选择一种作为第一拥塞控制规则。
在步骤502中，网络设备获取第一流量特征和第二流量特征，网络设备根据第一流量特征，获得第一回报值，根据第二流量特征，获得第二回报值。
网络设备获取第一流量特征和第二流量特征，第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，网络设备根据第一流量特征，获得第一回报值，第二流量特征包括第二设备在第一周期内根据第二拥塞控制参数转发流量时生成的统计信息，网络设备根据第二流量特征，获得第二回报值。获取回报值的方法与前述图3的步骤302中获取回报值的方法类似，具体此处不再赘述。
在步骤503中，网络设备获取第一设备的第一运行值和第二运行值，获取第二设备的第三运行值和第四运行值，确定差值的均值是否小于目标阈值。
网络设备获取第一设备的第一运行值和第二运行值，获取第二设备的第三运行值和第 四运行值，确定差值的均值是否小于目标阈值，差值的均值是指第一差值和第二差值的平均值，第一差值是指第一运行值和第二运行值的差值，第二差值是第三运行值和第四运行值的差值。
可选地，网络设备可以仅通过第一拥塞控制规则调整初始拥塞控制参数，仅通过第一拥塞控制规则调整初始拥塞控制规则是指，网络设备不利用步长调整网络设备根据第一拥塞控制规则得出的拥塞控制参数，网络设备直接将第一拥塞控制规则得出的拥塞控制参数作为第一设备控制流量转发的参数；在网络设备利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数之前，网络设备可以统计网络设备仅通过第一拥塞控制规则调整初始拥塞控制参数的连续调整次数。
关于运行值和目标阈值的描述与前述图3的步骤303中描述的类似，具体此处不再赘述。
在步骤504中、若差值的均值小于目标阈值，则网络设备利用第一步长正向修改第一拥塞控制参数，获得第三拥塞控制参数，网络设备利用第二步长正向修改第二拥塞控制参数，获得第四拥塞控制参数。
若差值的均值小于目标阈值，则网络设备利用第一步长正向修改第一拥塞控制参数，获得第三拥塞控制参数，网络设备利用第二步长正向修改第二拥塞控制参数，获得第四拥塞控制参数。
正向修改只是为了和后面描述的反向修改做区别，并不是指第一拥塞控制参数只能正向加的意思，例如，第一拥塞控制参数为100，第一步长为20，网络设备利用第一步长正向修改第一拥塞控制参数后，第二拥塞控制参数可以是80，也可以是120，为了方便理解本实施例的拥塞控制方法，本实施例以第二拥塞控制参数是120为例进行说明。
可选地，第一步长和第二步长为百分比值。
可选地，目标阈值可以根据第一设备的网络波动大小进行调节，当第一设备的网络波动较小时，即第一设备的流量变化较小，网络设备将目标阈值调小，当第一设备的网络波动较大时，即第一设备的流量变化较大，网络设备将目标阈值调大。
可选地，当网络设备统计网络设备仅通过第一拥塞控制规则调整初始拥塞控制参数的连续调整次数时，网络设备可以不获取第一设备的第一运行值和第二运行值，网络设备可以确定该连续调整次数是否大于设定阈值N，当网络设备确定该连续调整次数大于设定阈值N时，则网络设备执行利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数的步骤。
可选地，当网络设备既统计网络设备仅通过第一拥塞控制规则调整初始拥塞控制参数的连续调整次数，网络设备又获取了第一运行值，第二运行值，第三运行值以及第四运行值时，只要有一个条件满足，则网络设备执行利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数的步骤，即包括以下几种情况：
当网络设备确定第一运行值和第二运行值的差值大于目标阈值，网络设备确定该连续调整次数大于设定阈值N时，网络设备执行利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数的步骤。
当网络设备确定第一运行值和第二运行值的差值小于目标阈值，网络设备确定该连续调整次数小于设定阈值N时，网络设备执行利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数的步骤。
当网络设备确定第一运行值和第二运行值的差值小于目标阈值，网络设备确定该连续调整次数大于设定阈值N时，网络设备执行利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数的步骤。
可选地，步骤503可以不执行，当不执行步骤503时，网络设备获取到第一拥塞控制参数后，网络设备可以无需确认差值的均值是否小于目标阈值，直接利用第一步长正向修改第一拥塞控制参数，获得第二拥塞控制参数，利用第二步长正向修改第二拥塞控制参数，获得第四拥塞控制参数。
在步骤505中，网络设备获取第三流量特征和第四流量特征，网络设备根据第三流量特征，获得第三回报值，网络设备根据第四流量特征，获得第四回报值。
网络设备利用第一步长正向修改第一拥塞控制参数，获得第三拥塞控制参数后，网络设备可以获取第三流量特征，第三流量特征包括所述第一设备在第二周期内根据第三拥塞控制参数转发流量时生成的统计信息；第三流量特征具体可以是第一设备的统计信息，或第一设备的端口的统计信息，或第一设备的队列的统计信息；网络设备获取第三流量特征后，网络设备可以根据获取到的第三流量特征获得第三回报值，第三回报值的算法与前述图三的步骤302中第一回报值的算法类似。第三流量特征可以是第一设备在第二周期内某个瞬间的数值，也可以经过处理后的数值，例如，第三流量特征可以是第一设备在第二期内的平均值。
网络设备利用第二步长正向修改第二拥塞控制参数，获得第四拥塞控制参数后，网络设备可以获取第四流量特征，第四流量特征包括所述第二设备在第二周期内根据第四拥塞控制参数转发流量时生成的统计信息；第四流量特征具体可以是第二设备的统计信息，或第二设备的端口的统计信息，或第二设备的队列的统计信息；网络设备获取第四流量特征后，网络设备可以根据获取到的第四流量特征获得第四回报值，第四回报值的算法与前述图三的步骤302中第一回报值的算法类似。第四流量特征可以是第二设备在第二周期内某个瞬间的数值，也可以经过处理后的数值，例如，第四流量特征可以是第二设备在第二期内的平均值。
可选地，当网络设备是第一设备时，网络设备直接向转发芯片下发第三拥塞控制参数，让转发芯片控制第一设备在第二周期转发的流量，网络设备向第二设备发送第四拥塞控制参数，用于让第二设备根据第四拥塞控制参数控制转发的流量。
可选地，当网络设备是第二设备时，网络设备直接向转发芯片下发第四拥塞控制参数，让转发芯片控制第二设备在第二周期转发的流量，网络设备向第一设备发送第三拥塞控制参数，用于让第三设备根据第三拥塞控制参数控制转发的流量。
可选地，当网络设备为除第一设备和第二以外的设备时，网络设备在获取到第三拥塞控制参数和第四拥塞控制参数后，网络设备会向第一设备发送第三拥塞控制参数，用于让第一设备利用第三拥塞控制参数来控制第一设备在第二周期转发的流量，网络设备还会向 第二设备发送第四拥塞控制参数，用于让第二设备利用第四拥塞控制参数来控制第二设备在第二周期转发的流量。
在步骤506中，网络设备确认第三回报值与第四回报值的和是否大于第一回报值与第二回报值的和。
在步骤507中，若第三回报值与第四回报值的和大于第一回报值与第二回报值的和，则网络设备利用第一步长修改第一拥塞控制规则，获得第三拥塞控制规则，利用第二步长修改第二拥塞控制规则，获得第四拥塞控制规则。
可选地，在网络设备确定第三回报值与第四回报值的和大于第一回报值与第二回报值的和时，网络设备先不用第一步长修改第一拥塞控制规则，获得第三拥塞控制规则，利用第二步长修改第二拥塞控制规则，获得第四拥塞控制规则的步骤，而是利用第三步长继续正向修改第三拥塞控制参数，获得第五拥塞控制参数，利用第四步长继续正向修改第四拥塞控制参数，获得第六拥塞控制参数；网络设备获取第五流量特征，第五流量特征包括所述第一设备在第三周期内根据第五拥塞控制参数转发流量时生成的统计信息；网络设备获取第六流量特征，第六流量特征包括所述第二设备在第三周期内根据第六拥塞控制参数转发流量时生成的统计信息；网络设备根据第五流量特征，获得第五回报值；网络设备根据第六流量特征，获得第六回报值；若第五回报值与第六回报值的和大于第三回报值和第四回报值的和，则网络设备继续利用第五步长正向修改第五拥塞控制参数，获得第七拥塞控制参数，网络设备利用第六步长正向修改第六拥塞控制参数，获得第八拥塞控制参数，网络设备获取第七流量特征，第七流量特征包括所述第一设备在第四周期内根据第七拥塞控制参数转发流量时生成的统计信息；网络设备获取第八流量特征，第八流量特征包括所述第二设备在第四周期内根据第八拥塞控制参数转发流量时生成的统计信息；网络设备根据第七流量特征，获得第七回报值，网络设备根据第八流量特征，获得第八回报值。依次类推，直到第T+1回报值与第T+2回报值的和小于第T－1回报值与第T回报值的和，第T回报值对应第T步长，第T－1回报值对应第T－1步长，网络设备利用第一步长，第三步长，至第T－1步长中所有步长的和作为第一步长，利用第一步长修改第一拥塞控制规则，获得第三拥塞控制规则，网络设备利用第二步长，第四步长，至第T件步长中所有步长的和作为第二步长，利用第二步长修改第二拥塞控制规则，获得第四拥塞控制规则。
在步骤508中，若第三回报值与第四回报值的和小于第一回报值与第二回报值的和，则网络设备利用第五步长反向修改第一拥塞控制参数，获得第五拥塞控制参数，利用第六步长修改第一拥塞控制参数，获得第六拥塞控制参数。
在步骤509中，网络设备获取第五流量特征和第六流量特征，网络设备根据第五流量特征，获得第五回报值，网络设备根据第六流量特征，获得第六回报值。
网络设备利用第三步长反向修改第一拥塞控制参数，获得第五拥塞控制参数后，网络设备可以获取第五流量特征，第五流量特征包括所述第一设备在第三周期内根据第五拥塞控制参数转发流量时生成的统计信息；第五流量特征具体可以是第一设备的统计信息，或第一设备的端口的统计信息，或第一设备的队列的统计信息；网络设备获取第三流量特征后，网络设备可以根据获取到的第五流量特征获得第五回报值，第五回报值的算法与前述 图三的步骤302中第一回报值的算法类似。第五流量特征可以是第一设备在第三周期内某个瞬间的数值，也可以经过处理后的数值，例如，第五流量特征可以是第一设备在第三期内的平均值。
网络设备利用第四步长反向修改第二拥塞控制参数，获得第六拥塞控制参数后，网络设备可以获取第六流量特征，第六流量特征包括所述第二设备在第三周期内根据第六拥塞控制参数转发流量时生成的统计信息；第六流量特征具体可以是第二设备的统计信息，或第二设备的端口的统计信息，或第二设备的队列的统计信息；网络设备获取第六流量特征后，网络设备可以根据获取到的第六流量特征获得第六回报值，第六回报值的算法与前述图三的步骤302中第一回报值的算法类似。第六流量特征可以是第二设备在第三周期内某个瞬间的数值，也可以经过处理后的数值，例如，第六流量特征可以是第二设备在第三期内的平均值。
在步骤510中，若第五回报值与第六回报值的和大于第一回报值与第二回报值的和，则网络设备利用第五步长修改第一拥塞控制规则，获得第三拥塞控制规则，利用第六步长修改第二拥塞控制规则，获得第四拥塞控制规则。
可选地，与步骤507中的类似，在网络设备确定第五回报值与第六回报值的和大于第一回报值与第二回报值的和时，网络设备先不用第五步长修改第一拥塞控制规则，获得第三拥塞控制规则，利用第六步长修改第二拥塞控制规则，获得第四拥塞控制规则的步骤，而是利用第七步长继续反向修改第五拥塞控制参数，获得第七拥塞控制参数，利用第八步长继续反向修改第六拥塞控制参数，获得第八拥塞控制参数。
上面对本申请实施例中的拥塞控制方法进行了描述，下面对本申请实施例中的拥塞控制装置进行描述。
请参阅图6，为本申请提供的拥塞控制装置的一个实施例的结构示意图。
第一获取单元601，用于获取第一流量特征，第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，第一拥塞控制参数是根据第一拥塞控制规则获得的；
第二获取单元602，用于根据第一流量特征，获得第一回报值；
第三获取单元603，用于利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数；
第四获取单元604，用于获取第二流量特征，第二流量特征包括第一设备在第二周期内根据第二拥塞控制参数转发流量时生成的统计信息；
第五获取单元605，用于根据第二流量特征，获得第二回报值；
执行单元606，用于若第二回报值大于第一回报值，则执行相应的处理。
本实施例中，在经过第一拥塞控制规则获得第一拥塞控制参数后，第一获取单元601可以获取第一流量特征，该第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，第一拥塞控制参数是根据第一拥塞控制规则获得的，第二获取单元602根据第一流量特征可以获得第一回报值，在利用第一步长修改第一拥塞控制参数后，第三获取单元603，可以获得第二拥塞控制参数，第四获取单元604可以获取第 二流量特征，该第二流量特征包括第一设备在第二周期内根据第二拥塞控制参数转发流量时生成的统计信息，第五获取单元605根据该第二流量特征，可以获得第二回报值，若第二回报值大于第一回报值，则执行单元606执行相应的处理，其中，第三获取单元603利用第一步长修改第一拥塞控制参数，获得了可以获得更大回报值的第二拥塞控制参数，因为第二回报值大于第一回报值，因此第二拥塞控制参数优于第一拥塞控制参数，因此拥塞控制装置对第一拥塞控制规则的推理结果进行了优化，因此提升了第一拥塞控制规则的场景泛化性。
本实施例中，拥塞控制装置各单元所执行的操作与前述图2所示实施例中描述的类似，此处不再赘述。
请参阅图7，为本申请提供的拥塞控制装置的另一个实施例的结构示意图。
在前述图6的拥塞控制装置的基础上，本申请提供的拥塞控制装置还包括：
可选地，第一获取单元601还用于获取第一设备的第一运行值和第二运行值。
该拥塞控制装置还包括：
确定单元707，用于确定第一运行值和第二运行值的差值是否小于目标阈值；
第三获取单元603具体用于，若差值小于目标阈值，则执行利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数的步骤。
可选地，该拥塞控制装置还包括：
调整单元708，用于调整初始拥塞控制参数；
统计单元709，用于统计初始拥塞控制参数的连续调整次数；
第三获取单元603具体用于当连续调整次数大于设定阈值N时，执行利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数的步骤；
第一拥塞控制参数为将初始拥塞控制参数根据所述第一拥塞控制规则连续调整N次得到的。
可选地，第一公式包括：Q＝A×B。
可选地，第一拥塞控制规则为第一公式，第一公式为：
Q＝F(A，B)，其中，Q为拥塞控制参数，A为速率，B为设定时延，F(A，B)为跟A和B相关的函数。
可选地，第五获取单元605还用于利用第一步长修改第一公式，获得第二公式。
可选地，第五获取单元605还用于根据第二流量特征，利用第二公式获得第三拥塞控制参数；
第三拥塞控制参数用于第一设备控制转发的流量。
可选地，第一步长为百分比值。
可选地，第二公式为：
Q＝A×B×(1+C)，其中，C为第一步长。
可选地，执行单元606具体用于利用第二步长正向修改第二拥塞控制参数，获得第三拥塞控制参数；
执行单元606具体用于获取第三流量特征，第三流量特征包括第一设备在第三周期内 根据第三拥塞控制参数转发流量时生成的统计信息；
执行单元606具体用于根据第三流量特征获取第三回报值。
可选地，第三获取单元603还用于，若第三回报值小于第二回报值，利用第三步长反向修改第二拥塞控制参数，获得第四拥塞控制参数；
第四获取单元604还用于，获取第四流量特征，第四流量特征包括第一设备在第四周期内根据第四拥塞控制参数转发流量时生成的统计信息；
第五获取单元605还用于根据第四流量特征获取第四回报值。
可选地，该拥塞控制装置还包括：
修改单元710，用于若第四回报值大于第二回报值，利用第三步长和第一步长修改第一拥塞控制规则，获得第二拥塞控制规则。
可选地，该拥塞控制装置还包括：
生成单元711，用于根据第四流量特征，利用第二拥塞控制规则生成第一设备的新的拥塞控制参数，新的拥塞控制参数用于第一设备控制转发的流量。
可选地，第二步长大于第一步长。
可选地，第一获取单元601还用于获取第五流量特征，第五流量特征包括第二设备在第一周期内根据第五拥塞控制参数转发流量时生成的统计信息，第五拥塞控制参数是根据第三拥塞控制规则获得的；
第二获取单元602还用于根据第五流量特征，获得第五回报值；
第三获取单元603还用于利用第四步长修改第五拥塞控制参数，获得第六拥塞控制参数；
第四获取单元604还用于获取第六流量特征，第六流量特征包括第二设备在第二周期内根据第六拥塞控制参数转发流量时生成的统计信息；
第五获取单元605还用于根据第六流量特征，获得第六回报值；
执行单元606还用于若第六回报值与第二回报值之和大于第五回报值和第一回报值之和，则执行所述相应的处理。
可选地，执行单元706具体用于根据第四步长修改第三拥塞控制规则；
执行单元606具体用于根据第一步长修改第一拥塞控制规则。
可选地，执行单元706具体用于利用第二步长修改第二拥塞控制参数，获得第三拥塞控制参数；
执行单元606具体用于获取第三流量特征，第三流量特征包括第一设备在第三周期内根据第三拥塞控制参数转发流量时生成的统计信息；
执行单元606具体用于根据第三流量特征，获得第三回报值；
执行单元606具体用于利用第五步长修改第六拥塞控制参数，获得第七拥塞控制参数；
执行单元606具体用于获取第七流量特征，第七流量特征包括第二设备在第三周期内根据第七拥塞控制参数转发流量时生成的统计信息；
执行单元606具体用于根据第七流量特征，获得第七回报值。
本实施例中，拥塞控制装置各单元所执行的操作与前述图2和图3和图4所示实施例 中描述的类似，此处不再赘述。
上面对本申请实施例中的拥塞控制装置进行了描述，下面对本申请实施例中的拥塞控制设备进行描述。请参阅图8，为本申请提供的拥塞控制设备的一个实施例的结构示意图。
如图8所示，拥塞控制设备800包括处理器810，与所述处理器810耦接的存储器，通信接口830。拥塞控制设备800可以是图1的网络设备，第一设备或者第二设备。处理器810可以是中央处理器(central processing unit，CPU)，网络处理器(network processor，NP)或者CPU和NP的组合。处理器还可以是专用集成电路(application-specific integrated circuit，ASIC)，可编程逻辑器件(programmable logic device，PLD)或其组合。上述PLD可以是复杂可编程逻辑器件(complex programmable logic device，CPLD)，现场可编程逻辑门阵列(field-programmable gate array，FPGA)，通用阵列逻辑(generic array logic,GAL)或其任意组合。处理器810可以是指一个处理器，也可以包括多个处理器。存储器可以包括易失性存储器(volatile memory)，例如随机存取存储器(random access memory，RAM)；存储器也可以包括非易失性存储器(non-volatile memory)，例如只读存储器(read-only memory，ROM)，快闪存储器(flash memory)，硬盘(hard disk drive，HDD)或固态硬盘(solid-state drive，SSD)；存储器还可以包括上述种类的存储器的组合。存储器中存储有计算机可读指令，所述计算机可读指令包括多个软件模块，例如第一获取模块822，第二获取模块824，第三获取模块826，第四获取模块828，第五获取模块830，执行模块832。
处理器810执行各个软件模块后可以按照各个软件模块的指示进行相应的操作。在本实施例中，一个软件模块所执行的操作实际上是指处理器810根据所述软件模块的指示而执行的操作。
第一获取模块822可以用于获取第一流量特征，第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，第一拥塞控制参数是根据第一拥塞控制规则获得的。
第二获取模块824用于根据第一流量特征，获得第一回报值。
第三获取模块826用于利用第一步长修改第一拥塞控制参数，获得第二拥塞控制参数。
第四获取模块828用于获取第二流量特征，第二流量特征包括第一设备在第二周期内根据第二拥塞控制参数转发流量时生成的统计信息。
第五获取模块830用于根据第二流量特征，获得第二回报值。
执行模块832用于若第二回报值大于所述第一回报值，则执行相应的处理。
此外，处理器810执行存储器中的计算机可读指令后，可以按照所述计算机可读指令的指示，执行网络设备或第一设备或第二设备可以执行的全部操作，例如网络设备在与图2和图3和图4对应的实施例中执行的操作。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统，装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显 示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。

Claims (30)

  1. 一种拥塞控制方法，其特征在于，包括：
    获取第一流量特征，所述第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，所述第一拥塞控制参数是根据第一拥塞控制规则获得的；
    根据所述第一流量特征，获得第一回报值；
    利用第一步长修改所述第一拥塞控制参数，获得第二拥塞控制参数；
    获取第二流量特征，所述第二流量特征包括所述第一设备在第二周期内根据所述第二拥塞控制参数转发流量时生成的统计信息；
    根据所述第二流量特征，获得第二回报值；
    若所述第二回报值大于所述第一回报值，则执行相应的处理。
  2. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    获取所述第一设备的第一运行值和第二运行值；
    确定所述第一运行值和所述第二运行值的差值是否小于目标阈值；
    若所述差值小于所述目标阈值，则执行所述利用第一步长修改所述第一拥塞控制参数，获得第二拥塞控制参数的步骤。
  3. 根据权利要求1所述的方法，其特征在于，所述方法还包括：
    调整初始拥塞控制参数；
    统计所述初始拥塞控制参数的连续调整次数；
    当所述连续调整次数大于设定阈值N时，执行所述利用第一步长修改所述第一拥塞控制参数，获得第二拥塞控制参数的步骤；
    所述第一拥塞控制参数为将所述初始拥塞控制参数根据所述第一拥塞控制规则连续调整N+1次得到的。
  4. 根据权利要求1至3任意一项所述的方法，其特征在于，所述第一拥塞控制规则为第一公式，所述第一公式为：
    Q＝F(A，B)，其中，Q为拥塞控制参数，A为速率，B为设定时延，F(A，B)为跟A和B相关的函数。
  5. 根据权利要求4所述的方法，其特征在于，所述第一公式具体为：
    Q＝A×B。
  6. 根据权利要求5所述的方法，其特征在于，所述执行相应的处理包括：
    利用所述第一步长修改所述第一公式，获得第二公式。
  7. 根据权利要求6所述的方法，其特征在于，所述方法还包括：
    根据所述第二流量特征，利用所述第二公式获得第三拥塞控制参数；
    所述第三拥塞控制参数用于所述第一设备控制转发的流量。
  8. 根据权利要求7所述的方法，其特征在于，所述第一步长为百分比值。
  9. 根据权利要求6至8中任意一项所述的方法，其特征在于，所述第二公式为：
    Q＝A×B×(1+C)，其中，所述C为第一步长。
  10. 根据权利要求1至3任意一项所述的方法，其特征在于，
    所述执行相应的处理包括：
    利用第二步长正向修改所述第二拥塞控制参数，获得第三拥塞控制参数；
    获取第三流量特征，所述第三流量特征包括所述第一设备在第三周期内根据所述第三拥塞控制参数转发流量时生成的统计信息；
    根据所述第三流量特征获取第三回报值。
  11. 根据权利要求10所述的方法，其特征在于，所述方法还包括：
    若所述第三回报值小于所述第二回报值，利用第三步长反向修改所述第二拥塞控制参数，获得第四拥塞控制参数；
    获取第四流量特征，所述第四流量特征包括所述第一设备在第四周期内根据所述第四拥塞控制参数转发流量时生成的统计信息；
    根据所述第四流量特征获取第四回报值。
  12. 根据权利要求11所述的方法，其特征在于，所述方法还包括：
    若所述第四回报值大于所述第二回报值，利用第三步长和所述第一步长修改所述第一拥塞控制规则，获得第二拥塞控制规则。
  13. 根据权利要求12所述的方法，其特征在于，所述执行相应的处理之后，所述方法还包括：
    根据所述第四流量特征，利用所述第二拥塞控制规则生成所述第一设备的新的拥塞控制参数，所述新的拥塞控制参数用于所述第一设备控制转发的流量。
  14. 根据权利要求10至13任意一项所述的方法，其特征在于，所述第二步长大于所述第一步长。
  15. 根据权利要求1所述的方法，其特征在于，所述执行相应的处理包括：
    获取第五流量特征，所述第五流量特征包括第二设备在第一周期内根据第五拥塞控制参数转发流量时生成的统计信息，所述第五拥塞控制参数是根据第三拥塞控制规则获得的；
    根据所述第五流量特征，获得第五回报值；
    利用第四步长修改所述第五拥塞控制参数，获得第六拥塞控制参数；
    获取第六流量特征，所述第六流量特征包括所述第二设备在第二周期内根据所述第六拥塞控制参数转发流量时生成的统计信息；
    根据所述第六流量特征，获得第六回报值；
    若所述第六回报值与所述第二回报值之和大于所述第五回报值和所述第一回报值之和，则执行所述相应的处理。
  16. 根据权利要求15所述的方法，其特征在于，所述执行所述相应的处理包括：
    根据所述第四步长修改所述第三拥塞控制规则；
    根据所述第一步长修改所述第一拥塞控制规则。
  17. 根据权利要求15所述的方法，其特征在于，所述执行所述相应的处理包括：
    利用第二步长修改所述第二拥塞控制参数，获得第三拥塞控制参数；
    获取第三流量特征，所述第三流量特征包括所述第一设备在第三周期内根据所述第三拥塞控制参数转发流量时生成的统计信息；
    根据所述第三流量特征，获得第三回报值；
    利用第五步长修改所述第六拥塞控制参数，获得第七拥塞控制参数；
    获取第七流量特征，所述第七流量特征包括所述第二设备在第三周期内根据所述第七拥塞控制参数转发流量时生成的统计信息；
    根据所述第七流量特征，获得第七回报值。
  18. 一种拥塞控制装置，其特征在于，包括：
    第一获取单元，用于获取第一流量特征，所述第一流量特征包括第一设备在第一周期内根据第一拥塞控制参数转发流量时生成的统计信息，所述第一拥塞控制参数是根据第一拥塞控制规则获得的；
    第二获取单元，用于根据所述第一流量特征，获得第一回报值；
    第三获取单元，用于利用第一步长修改所述第一拥塞控制参数，获得第二拥塞控制参数；
    第四获取单元，用于获取第二流量特征，所述第二流量特征包括所述第一设备在第二周期内根据所述第二拥塞控制参数转发流量时生成的统计信息；
    第五获取单元，用于根据所述第二流量特征，获得第二回报值；
    执行单元，用于若所述第二回报值大于所述第一回报值，则执行相应的处理。
  19. 根据权利要求18所述的装置，其特征在于，
    所述第一获取单元还用于获取所述第一设备的第一运行值和第二运行值；
    所述装置还包括：
    确定单元，用于确定所述第一运行值和第二运行值的差值是否小于目标阈值；
    所述第三获取单元具体用于，若所述差值小于所述目标阈值，则执行所述利用第一步长修改所述第一拥塞控制参数，获得第二拥塞控制参数的步骤。
  20. 根据权利要求18所述的装置，其特征在于，所述装置还包括：
    调整单元，用于调整初始拥塞控制参数；
    统计单元，用于统计所述初始拥塞控制参数的连续调整次数；
    所述第三获取单元具体用于当所述连续调整次数大于设定阈值N时，执行所述利用第一步长修改所述第一拥塞控制参数，获得第二拥塞控制参数的步骤；
    所述第一拥塞控制参数为将所述初始拥塞控制参数根据所述第一拥塞控制规则连续调整N+1次得到的。
  21. 根据权利要求18至20任意一项所述的装置，其特征在于，所述第一拥塞控制规则为第一公式，所述第一公式为：
    Q＝F(A，B)，其中，Q为拥塞控制参数，A为速率，B为设定时延，F(A，B)为跟A和B相关的函数。
  22. 根据权利要求18至20任意一项所述的装置，其特征在于，
    所述执行单元具体用于利用第二步长正向修改所述第二拥塞控制参数，获得第三拥塞控制参数；
    所述执行单元具体用于获取第三流量特征；所述第三流量特征包括所述第一设备在第 三周期内根据所述第三拥塞控制参数转发流量时生成的统计信息；
    所述执行单元具体用于根据所述第三流量特征获取第三回报值。
  23. 根据权利要求22所述的装置，其特征在于，
    所述第三获取单元还用于，若所述第三回报值小于所述第二回报值，利用第三步长反向修改所述第二拥塞控制参数，获得第四拥塞控制参数；
    所述第四获取单元还用于，获取第四流量特征，所述第四流量特征包括所述第一设备在第四周期内根据所述第四拥塞控制参数转发流量时生成的统计信息；
    所述第五获取单元还用于根据所述第四流量特征获取第四回报值。
  24. 根据权利要求23所述的装置，其特征在于，所述装置还包括：
    修改单元，用于若所述第四回报值大于所述第二回报值，利用第三步长和所述第一步长修改所述第一拥塞控制规则，获得第二拥塞控制规则。
  25. 根据权利要求18至24任意一项所述的装置，其特征在于，
    所述装置还包括：
    生成单元，用于根据所述第四流量特征，利用所述第二拥塞控制规则生成所述第一设备的新的拥塞控制参数，所述新的拥塞控制参数用于所述第一设备控制转发的流量。
  26. 根据权利要求18至25中任意一项所述的装置，其特征在于，
    所述第一获取单元还用于获取第五流量特征，所述第五流量特征包括第二设备在第一周期内根据第五拥塞控制参数转发流量时生成的统计信息，所述第五拥塞控制参数是根据第三拥塞控制规则获得的；
    所述第一获取单元还用于根据所述第五流量特征，获得第五回报值；
    所述第三获取单元还用于利用第四步长修改所述第五拥塞控制参数，获得第六拥塞控制参数；
    所述第四获取单元还用于获取第六流量特征，所述第六流量特征包括所述第二设备在第二周期内根据所述第六拥塞控制参数转发流量时生成的统计信息；
    所述第五获取单元还用于根据所述第六流量特征，获得第六回报值；
    所述执行单元还用于若所述第六回报值与所述第二回报值之和大于所述第五回报值和所述第一回报值之和，则执行所述相应的处理。
  27. 根据权利要求26所述的装置，其特征在于，
    所述执行单元具体用于根据所述第四步长修改所述第三拥塞控制规则；
    所述执行单元具体用于根据所述第一步长修改所述第一拥塞控制规则。
  28. 一种拥塞控制设备，其特征在于，包括：存储器和处理器；
    其中，所述存储器用于存储程序；
    所述处理器用于执行所述存储器中的程序，包括执行如上述权利要求1至17中任意一项所述的方法。
  29. 一种计算机存储介质，其特征在于，所述计算机存储介质中存储有指令，所述指令在计算机上执行时，使得所述计算机执行如权利要求1至17中任一项所述的方法。
  30. 一种计算机程序产品，其特征在于，所述计算机程序产品在计算机上执行时，使 得所述计算机执行如权利要求1至17中任一项所述的方法。
PCT/CN2020/113775 2019-12-03 2020-09-07 拥塞控制方法以及相关设备 WO2021109661A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP20895560.9A EP4054134A4 (en) 2019-12-03 2020-09-07 SPACE MANAGEMENT METHOD AND ASSOCIATED DEVICE
US17/831,070 US20220294736A1 (en) 2019-12-03 2022-06-02 Congestion Control Method and Related Device

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911223339.7 2019-12-03
CN201911223339.7A CN112910789A (zh) 2019-12-03 2019-12-03 拥塞控制方法以及相关设备

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/831,070 Continuation US20220294736A1 (en) 2019-12-03 2022-06-02 Congestion Control Method and Related Device

Publications (1)

Publication Number Publication Date
WO2021109661A1 true WO2021109661A1 (zh) 2021-06-10

Family

ID=76104308

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/113775 WO2021109661A1 (zh) 2019-12-03 2020-09-07 拥塞控制方法以及相关设备

Country Status (4)

Country Link
US (1) US20220294736A1 (zh)
EP (1) EP4054134A4 (zh)
CN (1) CN112910789A (zh)
WO (1) WO2021109661A1 (zh)

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20210243125A1 (en) * 2020-01-31 2021-08-05 Hewlett Packard Enterprise Development Lp System and method for avoiding congestion in a computer network
CN113411263B (zh) * 2021-06-18 2023-03-14 中国工商银行股份有限公司 一种数据传输方法、装置、设备及存储介质
CN117544567B (zh) * 2024-01-09 2024-03-19 南京邮电大学 存传一体的rdma数据中心拥塞控制方法

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110149730A1 (en) * 2008-06-06 2011-06-23 Gabor Nemeth Technique for improving congestion control
WO2012097737A1 (zh) * 2011-01-18 2012-07-26 华为技术有限公司 一种数据传输控制方法及设备
US20140269326A1 (en) * 2013-03-15 2014-09-18 Google Inc. Dynamic congestion control
CN107079435A (zh) * 2015-05-26 2017-08-18 华为技术有限公司 数据传输装置、v2v通信系统及方法
WO2017161566A1 (zh) * 2016-03-25 2017-09-28 华为技术有限公司 一种拥塞控制方法、装置及相关设备
US10257097B1 (en) * 2016-10-04 2019-04-09 EMC IP Holding Company LLC Connection based selection of a network congestion control process

Family Cites Families (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7499453B2 (en) * 2000-05-19 2009-03-03 Cisco Technology, Inc. Apparatus and methods for incorporating bandwidth forecasting and dynamic bandwidth allocation into a broadband communication system
US8817615B2 (en) * 2010-03-05 2014-08-26 Nec Europe Ltd. Method for operating a wireless network and a wireless network
US20130155855A1 (en) * 2010-08-09 2013-06-20 Nokia Siemens Networks Oy Increasing Efficiency of Admission Control in a Network
US10484257B1 (en) * 2015-07-15 2019-11-19 Amazon Technologies, Inc. Network event automatic remediation service
CN108881056B (zh) * 2017-05-15 2022-02-25 华为技术有限公司 一种拥塞控制方法、网络设备及其网络接口控制器
US10708806B2 (en) * 2018-01-02 2020-07-07 Verizon Patent And Licensing Inc. Systems and methods for a self-organizing network based on user equipment information
CN110278157B (zh) * 2018-03-14 2022-08-09 华为技术有限公司 拥塞控制方法及网络设备
CN110120921B (zh) * 2019-05-13 2022-07-01 深圳市赛为智能股份有限公司 拥塞避免方法、装置、计算机设备及存储介质
JP7188609B2 (ja) * 2019-09-30 2022-12-13 日本電気株式会社 システム、方法及び制御装置

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20110149730A1 (en) * 2008-06-06 2011-06-23 Gabor Nemeth Technique for improving congestion control
WO2012097737A1 (zh) * 2011-01-18 2012-07-26 华为技术有限公司 一种数据传输控制方法及设备
US20140269326A1 (en) * 2013-03-15 2014-09-18 Google Inc. Dynamic congestion control
CN107079435A (zh) * 2015-05-26 2017-08-18 华为技术有限公司 数据传输装置、v2v通信系统及方法
WO2017161566A1 (zh) * 2016-03-25 2017-09-28 华为技术有限公司 一种拥塞控制方法、装置及相关设备
US10257097B1 (en) * 2016-10-04 2019-04-09 EMC IP Holding Company LLC Connection based selection of a network congestion control process

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
KONG, JINSHENG: "Intelligent Adaptive Control Approach of Congestion Control for Computer Networks", SYSTEMS ENGINEERING AND ELECTRONICS, vol. 27, no. 7, 1 July 2005 (2005-07-01), pages 1301 - 1303, XP055819294, ISSN: 1001-506x *
NTT DOCOMO, INC.: "Congestion Control for NR Sidelink", 3GPP DRAFT; R1-1906210 CONGESTION CONTROL, 3RD GENERATION PARTNERSHIP PROJECT (3GPP), MOBILE COMPETENCE CENTRE ; 650, ROUTE DES LUCIOLES ; F-06921 SOPHIA-ANTIPOLIS CEDEX ; FRANCE, vol. RAN WG1, no. Reno, USA; 20190513 - 20190517, 3 May 2019 (2019-05-03), Mobile Competence Centre ; 650, route des Lucioles ; F-06921 Sophia-Antipolis Cedex ; France, XP051708249 *
See also references of EP4054134A4

Also Published As

Publication number Publication date
EP4054134A1 (en) 2022-09-07
EP4054134A4 (en) 2023-01-11
US20220294736A1 (en) 2022-09-15
CN112910789A (zh) 2021-06-04

Similar Documents

Publication Publication Date Title
WO2021109661A1 (zh) 拥塞控制方法以及相关设备
KR102478440B1 (ko) 혼잡 제어 방법 및 네트워크 디바이스
JP7451689B2 (ja) ネットワーク輻輳処理方法、モデル更新方法、および関連装置
CN108768876B (zh) 一种面向机器学习框架的流量调度方法
US20220210026A1 (en) Network Parameter Configuration Method and Apparatus, Computer Device, and Storage Medium
US11888744B2 (en) Spin-leaf network congestion control method, node, system, and storage medium
CN112422443B (zh) 一种拥塞算法的自适应控制方法、存储介质、设备及系统
CN110351187B (zh) 数据中心网络中路径切换粒度自适应的负载均衡方法
CN102763385A (zh) 针对数据中心环境优化的通信输送
CN101969408B (zh) 基于分组dscp标记的主动队列管理方法
US20140211625A1 (en) Predictive management of a network buffer
CN112995048B (zh) 数据中心网络的阻塞控制与调度融合方法及终端设备
WO2015044719A1 (en) Apparatus for optimising a configuration of a communications network device
WO2020063299A1 (zh) 统计活跃流数目的方法、网络设备和系统
US20220200858A1 (en) Method and apparatus for configuring a network parameter
CN103888313A (zh) 一种预测流表项的最优超时时间的方法
CN115102905B (zh) 一种ecn水线调整方法及装置
CN109698925A (zh) 基于数据驱动的实时视频拥塞控制方法及装置
CN103269458A (zh) 一种针对窄带网络情况下视频传输的调节控制方法
Zhuo et al. Rack-level congestion control
Xia et al. A multi-objective reinforcement learning perspective on internet congestion control
CN106101005A (zh) 一种数据中心网络中基于分组长度的拥塞控制方法
CN109508433B (zh) 基于匹配算法性能调整的负载波动应对方法和系统
CN108737286B (zh) 一种全局拥塞价格的计算方法
CN118381765B (zh) 无损网络拥塞控制方法、装置、设备、介质及交换机系统

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20895560

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020895560

Country of ref document: EP

Effective date: 20220603

NENP Non-entry into the national phase

Ref country code: DE