CN114301845B - Self-adaptive data center network transmission protocol selection method - Google Patents
Self-adaptive data center network transmission protocol selection method Download PDFInfo
- Publication number
- CN114301845B CN114301845B CN202111628288.3A CN202111628288A CN114301845B CN 114301845 B CN114301845 B CN 114301845B CN 202111628288 A CN202111628288 A CN 202111628288A CN 114301845 B CN114301845 B CN 114301845B
- Authority
- CN
- China
- Prior art keywords
- threshold
- flow
- window
- rcc
- size
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
Abstract
The invention discloses a self-adaptive data center network transmission protocol selection system and a method, wherein the system comprises a stream information collector and a threshold learner; wherein: the flow information collector is used for collecting FCT and flow size information of the finished RCC flow of the passive congestion control scheme, and storing the FCT and the flow size information in the RCCHistor window; the information collected by the flow information collector is transmitted to the threshold learner for training the threshold learner; and the flow information collector is configured to collect scheduling delays of an active congestion control scheme PCC flow; the threshold learner is used for calculating the threshold online according to the information collected by the flow information collector. Compared with the prior art, the invention controls the queue length by transmitting the large flow by using the active congestion control, and directly transmits the small flow by using the passive congestion control, thereby avoiding overlong scheduling time and ensuring shorter queuing delay.
Description
Technical Field
The invention belongs to the technical field of flow control or congestion control, and particularly relates to a data center network transmission protocol selection method.
Background
With the growth of internet services such as real-time audio and video, electronic commerce, online games, stock exchanges, virtual reality and the like, the requirement on the time delay performance of a data center network is higher and higher, and the time delay performance is changed from the past second level to the present microsecond level. The time delay directly affects the satisfaction of the user and thus the revenue of the enterprise. Therefore, designing a low latency data center congestion control scheme becomes a hotspot problem in data center networks.
For congestion control problems in data center networks, a number of solutions have emerged in recent years. Passive congestion control (Reactive Congestion Control, RCC) schemes require probing the link conditions, blindly sending data into the network, and then adjusting the rate based on the congestion signal. For example, DCTCP adjusts the transmission window according to the proportion of packets marked ECN; the timer adjusts the transmission rate according to the precisely measured Round-Trip Time (RTT). But this is a post-regulation method, i.e. the situation that switch queues pile up, even packet loss, etc. already occurs during speed regulation, which seriously affects the delay of the network.
Active congestion control (Proactive Congestion Control, PCC) actively allocates bandwidth at the central controller or receiver based on global or local flow information. For example, fastpass decides an appropriate transmit time path for each packet at the central controller based on global flow information; the receiving end of ExpressPass sends out credit packets to the sending end according to the link condition, and the number of the credit packets is limited by utilizing the leaky bucket algorithm of the switch, so that the number of the data packets in the network is reversely controlled. But having a global or local view of the flow information is costly, requiring at least one RTT of schedule time overhead. The time of one RTT can be tolerated for large flows, but is fatal for small flows that account for the majority of data center networks, because small flows carry a small number of bytes and usually can be completed in one RTT, the flow completion time (Flow Completion Time, FCT) of a small flow will be held at least twice. This results in that small flows may run faster than PCC using RCC with the same flow distribution and load.
Disclosure of Invention
In view of the above-mentioned shortcomings in the prior art, the present invention is directed to a system and a method for adaptively selecting a data center network transmission protocol, which adaptively implements data center network transmission protocol selection by an algorithm when a stream is started.
The invention is realized by the following technical scheme:
an adaptive data center network transport protocol selection system includes a stream information collector and a threshold learner; wherein:
the flow information collector is used for collecting FCT and flow size information of the finished RCC flow of the passive congestion control scheme, and storing the FCT and the flow size information in the RCCHistor window; the information collected by the flow information collector is transmitted to the threshold learner for training the threshold learner; and the flow information collector is configured to collect scheduling delays of an active congestion control scheme PCC flow;
the threshold value learner is used for calculating a threshold value online according to the information collected by the flow information collector; the calculation process is as follows:
screening historical RCC flows with a flow completion time FCT less than the PCC leveling uniform scheduling delay d, namely RCCHistory, add to an empty set, if the size of the set is divided by the window, the ratio r w If the value is equal to 1, the last selected diversion threshold is too conservative or the network is not congested, and more RCC flows are sent to the network; if r w Greater than or equal to the preset value p and less than 1, indicating the last selected shunt threshold s t-1 Accurately, the current threshold s is output along with s t-1 And the largest stream size in the set increases or decreases; if the ratio r w Less than the preset value p and not equal to 1, indicating the last selected shunt threshold s t-1 Too large, resulting in excessive flows using RCCs, the network becomes congested, requiring a reduction in the shunt threshold to reduce the number of RCCs used; if no enough flows with FCT less than d are found in this window, the rcchisory window is doubled in size and the process is repeated again until the rcchisory window expands to the full rcchisory
An adaptive data center network transport protocol selection method, the method comprising the steps of:
step 1, in the initial stage, a sending end takes 10KB as an initial threshold value to shunt, if the flow size is larger than the threshold value, an active congestion control scheme is used, otherwise, the active congestion control scheme is used; recording the flow completion time FCT and flow size using DCTCP flow using rcchortry window, max window H size Setting 40MTU, setting current window w as initial window;
step 2, judging that the current window is larger than the maximum window? If yes, go to step 10; if not, turning to the step 3;
step 3, if not, screening out a historical RCC flow with FCT smaller than the PCC leveling uniform scheduling delay d, namely RCCHistor, from the current window w, and adding the historical RCC flow into an empty set;
step 4, judging the ratio r of the size of the set to the window w w Is greater than or equal to a preset value p? If yes, go to step 5;
step 5, judging the ratio r of the size of the set to the window w w Is equal to 1?
Step 6, if the size of the set is a proportion r of the window w w Equal to 1, then tableKnowing that the last selected split threshold is too conservative or the network is not congested allows more RCC flows to be sent into the network, the split threshold is set to:
threshold＝last_threshold+s
wherein threshold represents the shunt threshold, last_threshold represents the last shunt threshold selected last time, s is the current threshold increment;
step 7, if r w If the current threshold increment is greater than or equal to the preset value p and is not equal to 1, the last selected shunt threshold is accurate, the output current threshold increment is increased or decreased along with the last selected shunt threshold and the maximum flow size in the set, and the shunt threshold is expressed as follows:
threshold＝(MAX(set)+last_threshold)/2
where MAX (set) represents the maximum stream size;
step 8, outputting threshold;
step 9, if the ratio r w If the current window p is smaller than the preset value p, the current window p is enlarged to be twice as much as the original window p, namely w= 2*w;
step 10, setting the threshold to be the maximum value, i.e. threshold=max (set).
Compared with the prior art, the invention has the beneficial effects and advantages that:
the active congestion control is used for transmission of the large flow, so that the length of a queue is controlled, the passive congestion control is used for direct transmission of the small flow, the overlong scheduling time is avoided, and the experienced queuing delay is also ensured to be short.
Drawings
FIG. 1 is a schematic diagram of an adaptive data center network transport protocol selection system architecture according to the present invention;
FIG. 2 is a flowchart of the overall process of threshold learner calculation according to the present invention;
FIG. 3 is a graph showing the results of a test of the average delay of a adaptive data center network transport protocol selection method in combination with Fastpass according to the present invention;
fig. 4 is a test result of the tail delay of the adaptive data center network transport protocol selection method in combination with ExpressPass according to the present invention.
Detailed Description
The technical scheme of the invention is described in detail below with reference to the accompanying drawings and specific embodiments.
Fig. 1 is a schematic diagram of an adaptive data center network transport protocol selection system architecture according to the present invention. Including a flow information collector 10 and a threshold learner 20.
The flow information collector 10 is configured to collect information such as FCT and flow size of the RCC flow of the completed passive congestion control scheme, and store the information in the rcchortry window. If the new addition of a stream results in an RCCHistor window size greater than the maximum window, the first added stream is deleted to reduce program overhead. And, the flow information collector 10 is configured to collect scheduling delays of the active congestion control scheme PCC flows. The scheduling delay is equal to the time of the first packet transmission minus the flow start time. The above information collected by the flow information collector 10 is transmitted to the threshold learner 20 for training the threshold learner 20.
The threshold learner 20 calculates the threshold online from the information collected by the flow information collector 10, and finds the optimal shunt threshold by dynamically adapting to the network conditions. Whenever a new RCC flow is collected by the flow information collector 10, the threshold learner 20 is notified to update the threshold. In particular, when an RCC stream is generated, its size is directly compared to the current shunt threshold without waiting for the threshold learner to calculate. If the current RCC flow size is less than or equal to the split threshold, the RCC flow is sent directly into the network, otherwise the PCC flow is used to wait until the scheduling instruction is reissued, thereby protecting the small RCC flow from excessive scheduling time.
The calculation process of the threshold learner 20 is as follows:
first, a historical RCC flow, namely RCCHistor, with FCT (flow completion time ) less than PCC flow average scheduling delay d is screened out and added to an empty set, if the size of the set is divided byThe ratio r obtained by the window w Equal to 1, indicating that the last selected split threshold is too conservative or the network is not congested, more RCC flows can be sent into the network, so the MixCC of the present invention will heuristically raise the split threshold. If rw is greater than or equal to the preset value p and less than 1, indicating the last selected shunt threshold s t-1 Accurately, the current threshold s is output along with s t-1 And the largest stream size in the set increases or decreases; if the ratio r w Less than the preset value p and not equal to 1, indicating the last selected shunt threshold s t-1 Too large, resulting in excessive flows using RCCs, the network becomes congested, and the shunt threshold needs to be reduced to reduce the number of RCCs used. To improve efficiency, if the time required to search the entire rcchondristor window is too long, the threshold is first searched from a smaller rcchondristor window, and each time the rcchondristor window is searched, the latest RCC stream is searched. If no sufficient flows have been found in this window with FCTs less than d, the rcchisory window is doubled in size and the process is repeated again until the rcchisory window expands to the full rcchisory.
Fig. 2 is a flow chart of a method for selecting an adaptive data center network transmission protocol according to the present invention.
And step 1, in the initial stage, the sending end takes 10KB as an initial threshold value to shunt, if the flow size is larger than the threshold value, a Fastpass algorithm of an active congestion control scheme is used, and otherwise, DCTCP is used. FCT and flow size recording using DCTCP flows using RCCHistor window, maximum window H size Setting 40MTU (maximum transmission unit), setting current window w as initial window;
step 2, judging that the current window is larger than the maximum window? If yes, go to step 10; if not, turning to the step 3;
step 3, if not, a historical RCC flow, namely RCCHistor, with flow completion time FCT smaller than PCC leveling uniform scheduling delay d is screened out from the current window w and added into an empty set;
step 4, judging the ratio r of the size of the set to the window w w Is greater than or equal to a preset value p? If yes, go to step5；
Step 5, judging the ratio r of the size of the set to the window w w Is equal to 1?
Step 6, if the size of the set is a proportion r of the window w w Equal to 1, it indicates that the last selected split threshold is too conservative or the network is not congested, allowing more RCC flows to be sent into the network, so the MixCC of the present invention will heuristically raise the split threshold, setting the split threshold to:
threshold＝last_threshold+s
wherein threshold represents the shunt threshold, last_threshold represents the last shunt threshold selected last time, s is the current threshold increment;
step 7, if r w If the current threshold increment is greater than or equal to the preset value p and is not equal to 1, the last selected shunt threshold is accurate, the output current threshold increment is increased or decreased along with the last selected shunt threshold and the maximum flow size in the set, and the shunt threshold is expressed as follows:
threshold＝(MAX(set)+last_threshold)/2
where MAX (set) represents the maximum stream size;
step 8, outputting threshold;
step 9, if the ratio r w If the current window p is smaller than the preset value p, the current window p is enlarged to be twice as much as the original window p, namely w= 2*w;
in step 10, the threshold is set to a maximum value, i.e. threshold=max (set).
The embodiment of the invention is described as follows:
in the initial stage, the sending end takes 10KB as an initial threshold value to shunt, if the flow size is larger than the threshold value, the Fastpass algorithm of the active congestion control scheme is used, otherwise, the DCTCP algorithm of the active congestion control scheme is used. RCCHistor window records FCT and flow size using DCTCP flow, max Window H size Set to 40MTU (maximum transmission unit), when the number of records exceeds the maximum window, delete the current windowDCTCP flow with the forefront port. Setting initial Window init W The size is 10MTU. The scheduling delay of a stream using Fastpass, i.e., the time from the start of the stream to the time the first packet was sent into the network, is recorded. When the number of collected DCTCP flows reaches the maximum window, decision on the threshold is started. The decision process is as follows:
first, the window size H of RCCHistor is obtained size The current window size w is set to init W ；
And then, starting to circularly screen DCTCP streams with FCT smaller than d according to the current window size w, wherein the ending condition of the circulation is that the current window size exceeds the maximum window. The specific process is as follows: first stream H of each RCCHistor window added last from RCCHistor head Beginning, the w-th stream H to RCCHistor window head-w-1 Up to now, the flow with FCT smaller than d is selected from w DCTCP flows, and its flow size is added to an empty set.
Secondly, the proportion of set in the current window is denoted as r w Wherein r is w =size (set)/w. If r w Equal to 1, the output threshold s is equal to s t-1 Plus S, i.e. s=s t-1 +s, where S is set to the size of the maximum transmission unit; if r w P is greater than or equal to 1, s is equal to s t-1 Plus the average of the maximum stream size MAX (set) in the window, i.e. s= (MAX (set) +s t-1 ) 2; if r w Less than p, the current window p is doubled, w= 2*w.
Finally, when w exceeds H size When it is stated that the current window does not have enough DCTCP flows such that its FCT is less than d, the threshold needs to be reduced. The stream size of the largest DCTCP stream with FCT less than d is found as a new threshold, i.e. s=max (set), throughout the rcchisory window.
When the stream is started, comparing the size of the stream with a threshold value to judge which transmission protocol should be used; and once selected, the transport protocol is no longer changed, and the stream will always use the transport protocol it selected. The threshold calculator updates the threshold each time the rcchistore collects a certain number of newly completed DCTCP flows, otherwise it compares to the flow or the last threshold.
The invention carries out large-scale simulation experiments on YAPS. WebServer, cacheFollower, webSearch and DataMining real workloads were employed. The MixCC+Fastpass version is realized by utilizing YAPS, and the network topology is a leaf ridge topology and comprises 144 servers; the passive congestion control algorithm employed is the most common congestion control algorithm DCTCP for data centers.
FIG. 3 is a graph showing the results of a test of the average delay of the adaptive data center network transport protocol selection method in combination with Fastpass according to the present invention. From FIG. 3, it can be seen that the present invention (MixCC) significantly reduces the average FCT for small flows compared to both Fastpass and DCTCP. At WebServer, cacheFollower, webSearch and DataMining four workloads, mixCC reduces the average FCT of the streamlets by 67.6%, 72.7%, 45.5% and 43.9% compared to FastPass; the average FCT of the small flows was reduced by 95.8%, 98.7%, 82.1% and 97.5% compared to DCTCP.
As shown in fig. 4, the test result of the tail delay of the adaptive data center network transmission protocol selection method in combination with ExpressPass according to the present invention is shown. As can be seen from the figure, the present invention (MixCC) significantly reduces the P99 FCT of small flows compared to the prior art methods. Specifically, at most 71.1%, 69.8%, 63.9% and 53.9% lower P99 FCT for small flows compared to ExpressPass at WebServer, cacheFollower, webSearch and DataMining workloads; the P99 FCT of the small stream was reduced by at most 62.2%, 59.2%, 67.1% and 50.8% compared to Aeolus.
In summary, compared with the existing method, the method has better effect on the time delay performance of the small flow.
The foregoing is merely exemplary of the present application and is not intended to limit the present application. Various modifications and changes may be made to the present application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc. which are within the spirit and principles of the present application are intended to be included within the scope of the claims of the present application.
Claims (8)
1. An adaptive data center network transmission protocol selection system, comprising a stream information collector and a threshold learner; wherein:
the flow information collector is used for collecting FCT and flow size information of the finished RCC flow of the passive congestion control scheme, and storing the FCT and the flow size information in the RCCHistor window; the information collected by the flow information collector is transmitted to the threshold learner for training the threshold learner; and the flow information collector is configured to collect scheduling delays of an active congestion control scheme PCC flow;
the threshold value learner is used for calculating a threshold value online according to the information collected by the flow information collector; the calculation process is as follows:
the historical RCC flow, namely RCCHistor, with the filtered flow completion time FCT smaller than the PCC flow leveling average scheduling delay d is added into an empty set, and if the size of the set is divided by the ratio r obtained by a window w If the value is equal to 1, the last selected diversion threshold is too conservative or the network is not congested, and more RCC flows are sent to the network; if r w Greater than or equal to the preset value p and less than 1, indicating the last selected shunt threshold s t-1 Accurately, the current threshold s is output along with s t-1 And the largest stream size in the set increases or decreases; if the ratio r w Less than the preset value p and not equal to 1, indicating the last selected shunt threshold s t-1 Too large, resulting in excessive flows using RCCs, the network becomes congested, requiring a reduction in the shunt threshold to reduce the number of RCCs used; if no sufficient flows have been found in this window with FCTs less than d, the rcchisory window is doubled in size and the process is repeated again until the rcchisory window expands to the full rcchisory.
2. The adaptive data center network transport protocol selection system of claim 1, wherein if a new added stream results in an rcchisory window greater than a maximum window, the first added stream is deleted.
3. The adaptive data center network transport protocol selection system of claim 1, wherein each time a new RCC stream is collected by the stream information collector 10, the threshold learner 20 is notified to update the threshold, and when a RCC stream is generated, the size is directly compared with the current split threshold; if the current RCC flow size is less than or equal to the split threshold, the RCC flow is sent directly into the network, otherwise, the PCC flow is used to wait until the scheduling instruction is reissued, thereby freeing the small RCC flow from excessive scheduling time.
4. The adaptive data center network transport protocol selection system of claim 1, wherein if the time required to search the entire rcchondristor window is too long, the threshold is first searched from a smaller rcchondristor window, and each time the rcchondristor window is searched, the latest completed RCC stream is searched.
5. An adaptive data center network transport protocol selection method implemented based on an adaptive data center network transport protocol selection system as claimed in claim 1, the method comprising the steps of:
step 1, in an initial stage, a sending end takes 10KB as an initial threshold value to shunt, if the flow size is larger than the threshold value, an active congestion control scheme is used, otherwise, the active congestion control scheme is used; recording flow completion time FCT and flow size, maximum window H for flows using active congestion control scheme using rcchisory window size Setting 40MTU, setting current window w as initial window;
step 2, judging that the current window is larger than the maximum window; if yes, go to step 10; if not, turning to the step 3;
step 3, if not, screening out a historical RCC flow with FCT smaller than the PCC leveling uniform scheduling delay d, namely RCCHistor, from the current window w, and adding the historical RCC flow into an empty set;
step 4, judging the ratio r of the size of the set to the window w w Whether greater than or equal to a preset value p; if yes, go to step 5;
step 5, judging the setThe ratio r of the combined set size to the window w w Whether or not equal to 1;
step 6, if the size of the set is a proportion r of the window w w Equal to 1, the last selected offload threshold is too conservative or the network is not congested, allowing more RCC flows to be sent into the network, setting offload threshold to:
threshold＝last_threshold+s
wherein threshold represents the shunt threshold, last_threshold represents the last shunt threshold selected last time, s is the current threshold increment;
step 7, if r w If the current threshold increment is greater than or equal to the preset value p and is not equal to 1, the last selected shunt threshold is accurate, the output current threshold increment is increased or decreased along with the last selected shunt threshold and the maximum flow size in the set, and the shunt threshold is expressed as follows:
threshold＝(MAX(set)+last_threshold)/2
where MAX (set) represents the maximum stream size;
step 8, outputting threshold;
step 9, if the ratio r w If the current window p is smaller than the preset value p, the current window p is enlarged to be twice as much as the original window p, namely w= 2*w;
step 10, setting the threshold to be the maximum value, i.e. threshold=max (set).
6. The adaptive data center network transport protocol selection method of claim 5, wherein if a new added stream results in an rcchisory window greater than a maximum window, the first added stream is deleted.
7. The adaptive data center network transport protocol selection method as recited in claim 5, wherein each time a new RCC stream is collected by the stream information collector, the threshold learner 20 is notified to update the threshold, and when an RCC stream is generated, the size of the RCC stream is directly compared with the current split threshold; if the current RCC flow size is less than or equal to the split threshold, the RCC flow is sent directly into the network, otherwise, the PCC flow is used to wait until the scheduling instruction is reissued, thereby freeing the small RCC flow from excessive scheduling time.
8. The method of claim 5, wherein if the time required to search the entire rcchondristor window is too long, the threshold is first searched from a smaller rcchondristor window, and each time the rcchondristor window is searched, the latest RCC stream is searched.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111628288.3A CN114301845B (en) | 2021-12-28 | 2021-12-28 | Self-adaptive data center network transmission protocol selection method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111628288.3A CN114301845B (en) | 2021-12-28 | 2021-12-28 | Self-adaptive data center network transmission protocol selection method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114301845A CN114301845A (en) | 2022-04-08 |
| CN114301845B true CN114301845B (en) | 2023-08-01 |
Family
ID=80972574
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111628288.3A Active CN114301845B (en) | 2021-12-28 | 2021-12-28 | Self-adaptive data center network transmission protocol selection method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114301845B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116527585B (en) * | 2023-07-05 | 2023-08-29 | 天地信息网络研究院(安徽)有限公司 | Congestion control method for stream length perception |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016122225A1 (en) * | 2015-01-28 | 2016-08-04 | 삼성전자 주식회사 | Method and apparatus for controlling congestion in communication system |
| CN106533970A (en) * | 2016-11-02 | 2017-03-22 | 重庆大学 | Differential flow control method and device for cloud computing data center network |
| CN113543209A (en) * | 2021-06-24 | 2021-10-22 | 北京邮电大学 | Token scheduling-based congestion control method and device |
-
2021
- 2021-12-28 CN CN202111628288.3A patent/CN114301845B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2016122225A1 (en) * | 2015-01-28 | 2016-08-04 | 삼성전자 주식회사 | Method and apparatus for controlling congestion in communication system |
| CN106533970A (en) * | 2016-11-02 | 2017-03-22 | 重庆大学 | Differential flow control method and device for cloud computing data center network |
| CN113543209A (en) * | 2021-06-24 | 2021-10-22 | 北京邮电大学 | Token scheduling-based congestion control method and device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114301845A (en) | 2022-04-08 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| JP5010739B2 (en) | Method and system for aggregate bandwidth control | |
| KR100234212B1 (en) | The congestion control based on adaptive speed control in packet network | |
| US7142508B2 (en) | System and method for controlling data transfer rates on a network | |
| US9001663B2 (en) | Communication transport optimized for data center environment | |
| US6333917B1 (en) | Method and apparatus for red (random early detection) and enhancements. | |
| US5014265A (en) | Method and apparatus for congestion control in a data network | |
| US11558302B2 (en) | Data transmission method and apparatus | |
| JPH1065710A (en) | Digital network including mechanism for providing flow control information to transmission source | |
| WO2000024164A1 (en) | Method and apparatus for network control | |
| CN110730469B (en) | Method for predicting bandwidth based on extended Kalman wireless network and congestion control thereof | |
| CN110351187A (en) | Data center network Road diameter switches the adaptive load-balancing method of granularity | |
| CN113543209B (en) | Token scheduling-based congestion control method and device | |
| CN114301845B (en) | Self-adaptive data center network transmission protocol selection method | |
| CN109150756B (en) | Queue scheduling weight quantification method based on SDN power communication network | |
| CN112995048B (en) | Blocking control and scheduling fusion method of data center network and terminal equipment | |
| JP2003298640A (en) | Method and apparatus for controlling traffic flow rate | |
| CN111211988B (en) | Data transmission method and system for distributed machine learning | |
| JP3115554B2 (en) | ATM network traffic management system and method for controlling cell flow in an ATM network | |
| CN109688065B (en) | Parameter processing method and device and storage medium | |
| CN116389375A (en) | Network queue management method, device and router for live video stream | |
| CN113839880A (en) | Congestion control method in named data network | |
| Almomani et al. | Simulation Based Performance Evaluation of Several Active Queue Management Algorithms for Computer Network | |
| KR100907995B1 (en) | Network Link Capacity Design Apparatus for MPC-based Differentiation Service | |
| JP3790897B2 (en) | Packet transfer control system and method, program thereof, recording medium, and communication apparatus | |
| CN114553836B (en) | Data block transmission punctuality improving method based on reinforcement learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |