WO2019011015A1 - 一种进行业务调度的方法和装置 - Google Patents

一种进行业务调度的方法和装置 Download PDF

Info

Publication number
WO2019011015A1
WO2019011015A1 PCT/CN2018/081867 CN2018081867W WO2019011015A1 WO 2019011015 A1 WO2019011015 A1 WO 2019011015A1 CN 2018081867 W CN2018081867 W CN 2018081867W WO 2019011015 A1 WO2019011015 A1 WO 2019011015A1
Authority
WO
WIPO (PCT)
Prior art keywords
cdn
network
resource
service scheduling
scheduling
Prior art date
Application number
PCT/CN2018/081867
Other languages
English (en)
French (fr)
Inventor
赵瑞
Original Assignee
网宿科技股份有限公司
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 网宿科技股份有限公司 filed Critical 网宿科技股份有限公司
Priority to EP18832346.3A priority Critical patent/EP3629553B1/en
Publication of WO2019011015A1 publication Critical patent/WO2019011015A1/zh
Priority to US16/726,115 priority patent/US11128684B2/en

Links

Images

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/50—Network services
    • H04L67/60—Scheduling or organising the servicing of application requests, e.g. requests for application data transmissions using the analysis and optimisation of the required network resources
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L65/00—Network arrangements, protocols or services for supporting real-time applications in data packet communication
    • H04L65/60—Network streaming of media packets
    • H04L65/61—Network streaming of media packets for supporting one-way streaming services, e.g. Internet radio
    • H04L65/612—Network streaming of media packets for supporting one-way streaming services, e.g. Internet radio for unicast
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
    • G06F7/58—Random or pseudo-random number generators
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/004—Artificial life, i.e. computing arrangements simulating life
    • G06N3/006—Artificial life, i.e. computing arrangements simulating life based on simulated virtual individual or collective life forms, e.g. social simulations or particle swarm optimisation [PSO]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0803—Configuration setting
    • H04L41/0823—Configuration setting characterised by the purposes of a change of settings, e.g. optimising configuration for enhancing reliability
    • H04L41/083—Configuration setting characterised by the purposes of a change of settings, e.g. optimising configuration for enhancing reliability for increasing network speed
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/14—Network analysis or design
    • H04L41/145—Network analysis or design involving simulating, designing, planning or modelling of a network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/16—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks using machine learning or artificial intelligence
    • 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
    • H04L43/0805—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters by checking availability
    • H04L43/0817—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters by checking availability by checking functioning
    • 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
    • H04L43/091—Measuring contribution of individual network components to actual service level
    • 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
    • H04L43/50—Testing arrangements
    • H04L43/55—Testing of service level quality, e.g. simulating service usage
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/70—Admission control; Resource allocation
    • H04L47/80—Actions related to the user profile or the type of traffic
    • H04L47/803—Application aware
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L47/00—Traffic control in data switching networks
    • H04L47/70—Admission control; Resource allocation
    • H04L47/83—Admission control; Resource allocation based on usage prediction
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/06—Protocols specially adapted for file transfer, e.g. file transfer protocol [FTP]
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/01—Protocols
    • H04L67/10—Protocols in which an application is distributed across nodes in the network
    • H04L67/1001—Protocols in which an application is distributed across nodes in the network for accessing one among a plurality of replicated servers
    • H04L67/1004—Server selection for load balancing
    • H04L67/101—Server selection for load balancing based on network conditions
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/2866—Architectures; Arrangements
    • H04L67/289—Intermediate processing functionally located close to the data consumer application, e.g. in same machine, in same home or in same sub-network
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/50—Network services
    • H04L67/60—Scheduling or organising the servicing of application requests, e.g. requests for application data transmissions using the analysis and optimisation of the required network resources
    • H04L67/61—Scheduling or organising the servicing of application requests, e.g. requests for application data transmissions using the analysis and optimisation of the required network resources taking into account QoS or priority requirements
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L67/00—Network arrangements or protocols for supporting network services or applications
    • H04L67/50—Network services
    • H04L67/60—Scheduling or organising the servicing of application requests, e.g. requests for application data transmissions using the analysis and optimisation of the required network resources
    • H04L67/62—Establishing a time schedule for servicing the requests
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/06—Management of faults, events, alarms or notifications

Definitions

  • the present application relates to the field of network data transmission, and in particular, to a method and apparatus for performing service scheduling.
  • a CDN Content Delivery Network
  • a node server which can be called a network node.
  • the user can obtain the required content from the network node, thereby alleviating the network congestion and improving the speed of the user accessing the website.
  • the technician When selecting alternative resources, the technician needs to manually calculate the configuration status of the network resources, such as the load status, network status, physical location, service redundancy, etc. of each network node, and then analyze and process the configuration status to select Optimal alternative resources, so that business scheduling is less efficient due to the large amount of manual involvement required.
  • the configuration status of the network resources such as the load status, network status, physical location, service redundancy, etc.
  • the embodiment of the present application provides a method and apparatus for performing service scheduling.
  • the technical solution is as follows:
  • a method for performing service scheduling comprising:
  • the method further includes:
  • the initial service scheduling model is trained to adjust parameters of the service scheduling model.
  • the initial service scheduling model is trained based on the second configuration status, and the parameters of the service scheduling model are adjusted, including:
  • the third configuration status of the network resource in the CDN test environment after the service scheduling is scored by using a preset network scoring standard
  • the parameters of the service scheduling model are adjusted by a back propagation algorithm.
  • the method further includes:
  • the fourth configuration status of the network resources in the CDN after the service scheduling is scored by using a preset network scoring standard
  • the parameters of the service scheduling model are adjusted by a back propagation algorithm.
  • the current first configuration status of the network resource in the CDN is obtained, including:
  • the failure alarm model obtained by the training is used to determine a scheduling trigger probability
  • a random number between 0 and 1 is generated.
  • the random number is less than or equal to the scheduling trigger probability, the current first configuration status of the network resource in the CDN network is obtained.
  • the method further includes:
  • the fault alarm model is built in the form of a Markov chain.
  • an apparatus for performing service scheduling comprising:
  • An acquiring module configured to acquire a current first configuration status of a network resource in a CDN network when a preset scheduling trigger event is detected;
  • a generating module configured to generate, according to the first configuration status, a service resource scheduling model obtained by training, where the substitute resource list records at least one substitute resource and a weight corresponding to the at least one substitute resource;
  • a selection module configured to select, according to a preset first selection policy, a first replacement resource in the substitute resource list according to a weight corresponding to the at least one substitute resource
  • a scheduling module configured to perform service scheduling in the CDN entire network based on the first alternative resource.
  • the acquiring module is further configured to: when a preset scheduling trigger event is detected in the CDN test environment, obtain a current second configuration status of the network resource in the CDN test environment, where the CDN is The test environment is the CDN local network and/or the CDN virtual whole network;
  • the apparatus further includes a training module, configured to train an initial service scheduling model based on the second configuration condition, and adjust parameters of the service scheduling model.
  • the training module is specifically configured to:
  • the third configuration status of the network resource in the CDN test environment after the service scheduling is scored by using a preset network scoring standard
  • the parameters of the service scheduling model are adjusted by a back propagation algorithm.
  • the device further includes:
  • a scoring module configured to score a fourth configuration status of the network resources in the CDN after the service is scheduled, after the service is scheduled in the CDN, by using a preset network scoring standard
  • the adjusting module is configured to adjust parameters of the service scheduling model by using a back propagation algorithm according to the scoring result.
  • the acquiring module is specifically configured to:
  • the failure alarm model obtained by the training is used to determine a scheduling trigger probability
  • a random number between 0 and 1 is generated.
  • the random number is less than or equal to the scheduling trigger probability, the current first configuration status of the network resource in the CDN network is obtained.
  • the device further includes:
  • a module is configured to calculate a fault alarm model in the form of a Markov chain according to the fault alarm history data of the CDN.
  • a management device comprising a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or a set of instructions, the at least one instruction, the at least one instruction A program, the set of codes, or a set of instructions is loaded and executed by the processor to implement a method of performing service scheduling as described in the first aspect.
  • a fourth aspect provides a computer readable storage medium, where the storage medium stores at least one instruction, at least one program, a code set, or a set of instructions, the at least one instruction, the at least one program, and the code
  • the set or set of instructions is loaded and executed by the processor to implement the method of performing traffic scheduling as described in the first aspect.
  • the current first configuration status of the network resource in the CDN is obtained; and the service scheduling model obtained by the training is generated according to the first configuration status, and the alternative resource list is generated.
  • the substitute resource list records the weight corresponding to the at least one substitute resource and the at least one substitute resource; according to the preset first picking policy, selecting the first substitute resource in the substitute resource list according to the weight corresponding to the at least one substitute resource; Based on the first alternative resource, the service scheduling is performed in the entire CDN network.
  • the management device can understand and master how to select alternative resources, avoiding the inaccuracy, maintenance difficulty, and cost fluctuation caused by the artificially designated scheduling rules.
  • the management device directly selects the optimal alternative resource through the service scheduling model, and does not require a large amount of manual participation, so the efficiency of the service scheduling can be improved.
  • FIG. 1 is a flowchart of a method for performing service scheduling according to an embodiment of the present application
  • FIG. 2 is a flowchart of a method for training a service scheduling model according to an embodiment of the present application
  • FIG. 3 is a schematic diagram of a deep reinforcement learning network architecture provided by an embodiment of the present application.
  • FIG. 4 is a schematic structural diagram of an apparatus for performing service scheduling according to an embodiment of the present application.
  • FIG. 5 is a schematic structural diagram of an apparatus for performing service scheduling according to an embodiment of the present application.
  • FIG. 6 is a schematic structural diagram of an apparatus for performing service scheduling according to an embodiment of the present application.
  • FIG. 7 is a schematic structural diagram of an apparatus for performing service scheduling according to an embodiment of the present disclosure.
  • FIG. 8 is a schematic structural diagram of a management device according to an embodiment of the present application.
  • An embodiment of the present application provides a method for performing service scheduling, where the execution entity of the method is a CDN network-wide management device, and the management device is configured to monitor the configuration status of the network resources of the CDN and the CDN based on the configuration status.
  • the network service in the network is scheduled.
  • the management device may be a separate server, dedicated to management, or a node server supporting the network service in the CDN network.
  • the function of the management device in this embodiment may be implemented by an independent device or by a device group composed of multiple devices.
  • the foregoing management device may include a processor, a memory, and a transceiver.
  • the processor may be configured to perform processing for performing service scheduling in the following process
  • the memory may be used to store data and generated data in the following processing process, and send and receive the data.
  • the device can be used to receive and transmit relevant data in the following processing.
  • the management device is an independent management device of the CDN network as an example, and other situations are similar.
  • Step 101 When a preset scheduling trigger event is detected, obtain a current first configuration status of the network resource in the CDN network.
  • the CDN network provides the entire resource environment of the CDN acceleration service on the network; the network resource may be the processing resource of the network node and/or the transmission resource of the transmission line, and the configuration status of the network resource may include the CPU usage of each network node, Service redundancy, input and output values, and bandwidth usage, response time, and packet loss rate of transmission lines between network nodes.
  • the technician on the CDN management side may set a scheduling trigger event for performing service scheduling in the management device in advance.
  • the scheduling trigger event may include a fault prediction alarm for the CDN network, a human-registered fault occurrence, and a timing scheduling task arrival trigger. Time and so on.
  • the management device can determine whether to perform service scheduling processing by monitoring whether the scheduling trigger event occurs in real time.
  • the preset scheduling trigger event is detected, the current first configuration status of the network resources in the CDN may be obtained first. Certainly, under certain circumstances, depending on the service scheduling rules, only the configuration status of some network resources associated with the service scheduling may be obtained in the CDN network.
  • the configuration status of the network node and the network nodes on the transmission line and the network resources in the vicinity of the network node, and the configuration status of the acquired network resources involved may be acquired, and the specific processing is performed. Keep it consistent and don't repeat them one by one.
  • the service scheduling may be triggered in advance by using a fault alarm model.
  • the processing of step 101 may be as follows: when detecting that the state parameter of the first network resource in the CDN is abnormal, the fault alarm model obtained through training, The scheduling trigger probability is determined; a random number between 0 and 1 is generated, and when the random number is less than or equal to the scheduling trigger probability, the current first configuration status of the network resource in the CDN network is obtained.
  • a fault alarm model for predicting a fault in the CDN network may be preset in the management device, and the fault alarm model may be used to predict the bottom machine index (ie, the state parameter of the network resource) in the CDN network. If a fault occurs, the probability of service scheduling needs to be performed. If the probability is high, the scheduling is triggered with a large probability.
  • the management device can obtain the state parameter of the network resource in the CDN.
  • the fault alarm model obtained by the training can determine the scheduling trigger probability. After that, a random number between 0 and 1 can be generated, and then the random number is compared with the scheduling trigger probability. When the random number is less than or equal to the scheduling trigger probability, the service scheduling is triggered, and then the management device can obtain the CDN.
  • the current first configuration status of network resources in the network can be specifically implemented by the following formula:
  • P is the scheduling trigger probability determined by the fault alarm model
  • ⁇ (t) is a randomly generated random number between 0 and 1 in the case of time t
  • the training process of the fault alarm model may be as follows: according to the fault alarm historical data of the CDN, the fault alarm model is established in the form of a Markov chain.
  • the management device can establish a fault alarm model according to the recent fault alarm history data in the CDN network, and continuously update the fault alarm model through the fault alarm data generated in real time in the CDN network.
  • the fault alarm history data is generated by monitoring each node server in the CDN network, from the state parameter exception to the historical data that actually triggers the service scheduling.
  • the management device can set the state parameter abnormality to the conversion rule that triggers the service scheduling based on the fault alarm history data (also referred to as the conversion rule of the underlying machine indicator abnormal alarm to the high-level alarm), for example, based on the time series, the high-level alarm is based on
  • the underlying indicator is generated by an abnormal alarm, and the high-level alarm is only related to its immediate next-level alarm. Therefore, the Markov chain can be used for calculation:
  • the Markov chain is a sequence of random variables with Markov properties, which describes a sequence of states, each state value depending on the previous limited state.
  • the scope of these random variables is a collection of all their possible values, called the "state space.”
  • the random variable X n represents the state at time n
  • X n+1 is only a function of X n .
  • the management device can calculate the fault alarm model in the form of a Markov chain according to the fault alarm history data of the CDN.
  • the association relationship between the state parameter anomalies and the triggering service scheduling is established, and in the process of detecting the state parameter changes, the subsequent faults are predicted, thereby achieving pre-scheduling and ensuring a certain degree of protection. Quality of service and user experience.
  • Step 102 Generate a substitute resource list by using the trained service scheduling model according to the first configuration status.
  • the substitute resource list records the weight corresponding to the at least one substitute resource and the at least one substitute resource.
  • the management device may pre-store a service scheduling model after a large amount of learning training, and the service scheduling model may be a deep reinforcement learning model established on a large-scale discrete state-action space, and iterative learning is performed through sample training.
  • Traditional reinforcement learning is an algorithm for solving the discrete state-action space problem, but when the discrete state and the action space are too large, traditional reinforcement learning is difficult to deal with these situations effectively.
  • the deep reinforcement learning model combines the two learning methods of reinforcement learning and deep learning, and can support the learning tasks on the large-scale discrete state-action space. It is suitable for the processing of business scheduling in this embodiment.
  • the specific model training can be See later.
  • the management device may obtain the state vector of the current network resource configuration by using the first configuration status, and then retrieve the service scheduling model obtained by the training, and use the status vector as the status vector.
  • the input of the service scheduling model is calculated by using the forward propagation algorithm of the hidden layer of the service scheduling model, and then the replacement resource list (also referred to as the alternative resource probability list) can be obtained by the output layer of the service scheduling model, and the record in the alternative resource list is recorded.
  • Step 103 Select, according to the preset first selection policy, the first alternative resource in the substitute resource list according to the weight corresponding to the at least one substitute resource.
  • the management device may select, according to the preset first selection policy, the weight corresponding to each of the substitute resources recorded in the substitute resource list, and select the foregoing alternative resource list.
  • the first alternative resource is to select the optimal alternative resource.
  • the first selection strategy is a selection strategy used when training the service scheduling model.
  • the strategy of the first selection strategy is not specifically defined, and the technical personnel on the management side may select a selection strategy according to actual conditions. For example, it may be a greedy strategy, that is, selecting the corresponding alternative resource with the highest weight as the first alternative resource in the alternative resource list, and first selecting the N alternative resources with the highest weight, and then randomly selecting the first An alternative resource.
  • Step 104 Perform service scheduling in the entire CDN network based on the first alternative resource.
  • the service scheduling process may be performed in the CDN entire network based on the first alternative resource, that is, the corresponding service is performed by using the first alternative resource.
  • the service scheduling model may be first adjusted in the CDN test environment, and the corresponding processing may be as follows: when the pre-detection is detected in the CDN test environment.
  • the scheduling trigger event occurs, the current second configuration status of the network resource in the CDN test environment is obtained; and based on the second configuration status, the initial service scheduling model is trained to adjust the parameters of the service scheduling model.
  • the CDN test environment is a CDN local network and/or a CDN virtual whole network.
  • the management device may construct a CDN test environment for training the service scheduling model.
  • the initial service scheduling model may be iteratively trained based on various scheduling trigger events to obtain a more mature service.
  • Scheduling model The specific processing may be as follows. When a preset scheduling trigger event is detected in the CDN test environment, the management device may acquire the current second configuration status of the network resource in the CDN test environment, and then perform initial service based on the second configuration status. The scheduling model is trained to adjust the parameters of the business scheduling model.
  • the above CDN test environment may be a CDN local network and/or a CDN virtual whole network.
  • the purpose of adopting the CDN local network is to plan a small-scale test environment from the CDN network environment, so that the negative impact of the service scheduling model during training can be controlled in the local network.
  • the local network may be divided according to the physical location of the network node, the operator to which the service belongs, or the user level.
  • the CDN virtual whole network may refer to a network environment that is virtualized according to the real CDN network.
  • the network environment supports the service scheduling system to interact with it, and supports the response of the real CDN network to the scheduling result according to the scheduling result.
  • the scheduling trigger event can be simply simulated by the cellular automaton model.
  • the training service scheduling model based on the CDN virtual whole network has the following advantages: First, the CDN local network belongs to the real network environment, and it is necessary to wait for the scheduling trigger event to appear before the service scheduling. Scheduling processing is performed to accumulate training data. In this way, model training requires a relatively long period of time. For the CDN virtual whole network, the simulation of the scheduling trigger event can be performed by the cellular automaton, so that the time period of the model training can be greatly shortened.
  • the CDN local network is to reduce the negative impact of the service scheduling model on the online service quality and cost during the training period, and a subset environment planned in the CDN network, although the subset environment Being able to limit negative impacts to a small extent, but not completely eliminating negative effects.
  • the data generated for model training in the subset environment is a local data set, which may be very different from the global data set of the CDN network.
  • the service scheduling model generated by the training may be over-fitting and cannot be generalized to the environment of the CDN network. Any processing in the CDN virtual network will not affect the real CDN network, and the CDN virtual network simulates the real CDN network environment, which can effectively avoid the limitations of the data layer.
  • the training process of the foregoing service scheduling model may specifically include the following steps as shown in FIG. 2:
  • Step 201 Generate an alternative resource list by using an initial service scheduling model based on the second configuration status.
  • the management device may obtain a state vector of the current network-wide resource configuration by using the second configuration status, and then retrieve an initial service scheduling model, and The state vector is used as an input of the initial service scheduling model, and is calculated by using a forward layer propagation algorithm of the hidden layer of the service scheduling model, and then an alternative resource list can be obtained by the output layer of the service scheduling model, where at least the recorded resource list is recorded.
  • An alternative resource and a weight corresponding to at least one alternative resource is recorded.
  • the structure of the service scheduling model can refer to the deep reinforcement learning network architecture shown in FIG. 3, wherein the input layer is a state vector, the output layer is a weight of an alternative resource, and the hidden layer is a linear or nonlinear calculation of a neuron.
  • the business scheduling model can be based on the initialization parameters, according to the state vector input by the input layer, through the hidden layer forward propagation calculation, so that the output layer output weights, the corresponding calculation formula can be as follows:
  • Representing the l+1th unit i-th unit input weighted sum, x 1 , x 2 , ..., x n are input units, Is the bias unit of the i-th unit of the first layer, It is the connection weight between the jth unit of the 1st layer and the i-th unit of the l+1th layer.
  • f( ⁇ ) is an activation function, which may be a function such as sigmoid or tanh.
  • Step 202 Select a second alternative resource in the substitute resource list according to the first selection policy, and perform service scheduling in the CDN test environment based on the second alternative resource.
  • the management device may select the second in the alternative resource list according to the weight corresponding to each alternative resource recorded in the substitute resource list according to the first selection policy. Alternative resources. Then, the management device can perform traffic scheduling based on the second alternative resource in the CDN test environment. It is worth mentioning that, in the training phase of the service scheduling model, the accuracy of the weights of the alternative resources in the above alternative resource list cannot be guaranteed, so the second alternative resource selected may not be the actual optimal replacement resource.
  • Step 203 The third configuration status of the network resource in the CDN test environment is scored by the preset network scoring standard.
  • the technical personnel on the management side can comprehensively consider the multi-dimensional data such as the bandwidth condition, the service capability, the user coverage, and the response time of the network node, establish a network quality scoring standard, and also can plan the network node, Multi-dimensional data such as billing type and billing coefficient are comprehensively considered to establish a network cost scoring standard.
  • quality scores can be firstly obtained from the perspective of network resources, such as quality scores of individual node servers, service cluster quality scores, etc., and quality scores can be performed from the perspective of users, such as user response time, etc., and then the quality of the two dimensions.
  • the score is comprehensively analyzed; then, based on the billing unit (such as a single network node), the cost score generated by the current CDN is calculated according to different billing types and billing coefficients, and finally the quality score and the cost score can be integrated. Generate a network rating standard and store the network rating criteria in the management device.
  • the management device may score the third configuration status of the network resource in the scheduled CDN test environment by using the preset network scoring standard.
  • Step 204 Adjust the parameters of the service scheduling model by using a back propagation algorithm according to the scoring result.
  • the parameters of the service scheduling model may be adjusted by using a back propagation algorithm.
  • the back propagation algorithm can be as follows:
  • Equation 1 calculating the error produced by the last layer of the business scheduling model:
  • Equation 2 calculating the error produced by each layer of the business scheduling model:
  • Equation 3 calculate the gradient of the weight:
  • Equation 4 calculate the gradient of the bias:
  • L represents the last layer of the hidden layer
  • C represents the true error between the model output value and the standard value
  • ⁇ L represents the error of the last layer calculated
  • ⁇ l represents the calculated error of the first layer
  • ⁇ For the Hadamard product Represents the derivative of the last layer of error for the output of the last layer of cells
  • ⁇ '(z L ) represents the derivative of the activation function for the last layer of unit inputs
  • ⁇ ′(z l ) represents the derivative of the activation function for the input of the layer 1 unit .
  • step 104 may score the scheduling result, and then adjust the parameters based on the scoring result, and the corresponding processing may be as follows: by using a preset network scoring standard, After the service scheduling, the fourth configuration status of the network resources in the CDN network is scored; according to the scoring result, the parameters of the service scheduling model are adjusted by a back propagation algorithm.
  • the management device can obtain the fourth configuration status of the network resources in the CDN network after the service scheduling, and then can adopt a preset network scoring standard. The fourth configuration status is scored. After that, the parameters of the business scheduling model can be adjusted according to the scoring result through the back propagation algorithm.
  • the service scheduling model utilizes the high abstraction capability of the deep neural network, which can effectively characterize the complex problems in the CDN network environment, and update the parameters in the model through the back propagation algorithm according to the feedback of the network scoring standard.
  • the business scheduling problem can be comprehensively considered and the global optimal solution can be selected.
  • the configuration status of the network resource in the CDN test environment is set as the input of the model, and the model outputs the alternative resource list based on the current parameter, and then selects the alternative resource according to a certain selection strategy, and then The network scoring standard is used to evaluate the service status of the alternative resource and feed back to the model.
  • the model After receiving the feedback, the model adjusts the model parameters according to the back propagation algorithm, so that the subsequent scheduling can proceed toward the global optimal direction.
  • a better performance business scheduling model can be generated.
  • the service scheduling model can be applied to the CDN network environment, and learning continues during subsequent operations.
  • the service scheduling model can select alternative resources that are expected to have a positive impact on the future. The business is performed, and the selection of the alternative resources can be evaluated through the network scoring standard, so that the business scheduling model can further adjust the parameters according to the scoring results.
  • the current first configuration status of the network resource in the CDN is obtained; and the service scheduling model obtained by the training is generated according to the first configuration status, and the alternative resource list is generated.
  • the substitute resource list records the weight corresponding to the at least one substitute resource and the at least one substitute resource; according to the preset first picking policy, selecting the first substitute resource in the substitute resource list according to the weight corresponding to the at least one substitute resource; Based on the first alternative resource, the service scheduling is performed in the entire CDN network.
  • the management device directly selects the optimal alternative resource through the service scheduling model, and does not require a large amount of manual participation, so the efficiency of the service scheduling can be improved.
  • the embodiment of the present application further provides a device for performing service scheduling.
  • the device includes:
  • the obtaining module 401 is configured to obtain a current first configuration status of the network resource in the CDN network when the preset scheduling trigger event is detected;
  • the generating module 402 is configured to generate, according to the first configuration status, a substitute resource list by using the trained service scheduling model, where the substitute resource list records at least one substitute resource and a weight corresponding to the at least one substitute resource ;
  • the selecting module 403 is configured to select, according to the preset first selection policy, the first alternative resource in the substitute resource list according to the weight corresponding to the at least one substitute resource;
  • the scheduling module 404 is configured to perform service scheduling in the CDN entire network based on the first alternative resource.
  • the obtaining module 401 is further configured to: when a preset scheduling trigger event is detected in the CDN test environment, obtain a current second configuration status of the network resource in the CDN test environment, where the The CDN test environment is the CDN local network and/or the CDN virtual whole network;
  • the apparatus further includes a training module 405, configured to train an initial service scheduling model based on the second configuration status, and adjust parameters of the service scheduling model.
  • the training module 405 is specifically configured to:
  • the third configuration status of the network resource in the CDN test environment after the service scheduling is scored by using a preset network scoring standard
  • the parameters of the service scheduling model are adjusted by a back propagation algorithm.
  • the apparatus further includes:
  • the scoring module 406 is configured to score the fourth configuration status of the network resources in the CDN after the service is scheduled, after the service scheduling in the CDN is performed by using a predetermined network scoring standard;
  • the adjusting module 407 is configured to adjust parameters of the service scheduling model by using a back propagation algorithm according to the scoring result.
  • the obtaining module 401 is specifically configured to:
  • the failure alarm model obtained by the training is used to determine a scheduling trigger probability
  • a random number between 0 and 1 is generated.
  • the random number is less than or equal to the scheduling trigger probability, the current first configuration status of the network resource in the CDN network is obtained.
  • the apparatus further includes:
  • the establishing module 408 is configured to calculate a fault alarm model in the form of a Markov chain according to the fault alarm history data of the CDN.
  • the current first configuration status of the network resource in the CDN is obtained; and the service scheduling model obtained by the training is generated according to the first configuration status, and the alternative resource list is generated.
  • the substitute resource list records the weight corresponding to the at least one substitute resource and the at least one substitute resource; according to the preset first picking policy, selecting the first substitute resource in the substitute resource list according to the weight corresponding to the at least one substitute resource; Based on the first alternative resource, the service scheduling is performed in the entire CDN network.
  • the management device directly selects the optimal alternative resource through the service scheduling model, and does not require a large amount of manual participation, so the efficiency of the service scheduling can be improved.
  • the apparatus for performing service scheduling in the foregoing embodiment is only illustrated by dividing the foregoing functional modules. In an actual application, the foregoing functions may be allocated by different functional modules according to requirements. Upon completion, the internal structure of the device is divided into different functional modules to perform all or part of the functions described above.
  • the apparatus for performing service scheduling provided by the foregoing embodiment is the same as the method for performing the service scheduling. The specific implementation process is described in detail in the method embodiment, and details are not described herein again.
  • FIG. 8 is a schematic structural diagram of a management device according to an embodiment of the present application.
  • the management device 800 can vary considerably depending on configuration or performance, and can include one or more central processors 822 (eg, one or more processors) and memory 832, one or more storage applications 842 or Storage medium 830 of data 844 (eg, one or one storage device in Shanghai).
  • the memory 832 and the storage medium 830 may be short-term storage or persistent storage.
  • the program stored on storage medium 830 may include one or more modules (not shown), each of which may include a series of instruction operations in the server.
  • central processor 822 can be configured to communicate with storage medium 830, executing a series of instruction operations in storage medium 830 on management device 800.
  • Management device 800 may also include one or more power supplies 826, one or more wired or wireless network interfaces 850, one or more input and output interfaces 858, one or more keyboards 856, and/or one or more operating systems. 841, such as WindowsServerTM, MacOSXTM, UnixTM, LinuxTM, FreeBSDTM, etc.
  • Management device 800 can include a memory, and one or more programs, wherein one or more programs are stored in a memory and configured to be executed by one or more processors to execute the one or more programs for execution The above instructions for performing business scheduling.
  • a person skilled in the art may understand that all or part of the steps of implementing the above embodiments may be completed by hardware, or may be instructed by a program to execute related hardware, and the program may be stored in a computer readable storage medium.
  • the storage medium mentioned may be a read only memory, a magnetic disk or an optical disk or the like.

Landscapes

  • Engineering & Computer Science (AREA)
  • Computer Networks & Wireless Communication (AREA)
  • Signal Processing (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Molecular Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computational Mathematics (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Environmental & Geological Engineering (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Databases & Information Systems (AREA)
  • Probability & Statistics with Applications (AREA)
  • Algebra (AREA)
  • Quality & Reliability (AREA)
  • Medical Informatics (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

本申请公开了一种进行业务调度的方法和装置，属于网络数据传输领域。所述方法包括：当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况；根据第一配置状况，通过训练得到的业务调度模型，生成替代资源列表，替代资源列表中记录有至少一个替代资源和至少一个替代资源对应的权值；按照预设的第一挑选策略，根据至少一个替代资源对应的权值，在替代资源列表中选择第一替代资源；基于第一替代资源，在CDN全网中进行业务调度。采用本申请，可以提高业务调度的效率。

Description

一种进行业务调度的方法和装置 技术领域
本申请涉及网络数据传输领域，特别涉及一种进行业务调度的方法和装置。
背景技术
CDN(Content Delivery Network，内容分发网络)是一种在服务提供方和消费方之间，通过架设节点服务器(可称为网络节点)的网络。通过该网络，用户可以就近从网络节点处获取所需的内容，从而可以达到缓解网络拥塞，提高用户访问网站的速度的目的。
当某个网络节点在超负荷状态下持续运转，或者网络出现异常波动时，CDN管理侧的技术人员需要对CDN全网中的部分业务进行调度，即选择出较优的替代资源，将上述部分业务移至该替代资源上处理。
在实现本申请的过程中，发明人发现现有技术至少存在以下问题：
在选择替代资源时，技术人员需要先人工统计网络资源的配置状况，如各网络节点的负载情况、网络情况、物理位置、服务冗余度等，然后再对配置状况进行分析处理，以选择出最优的替代资源，这样，由于需要大量的人工参与，业务调度的效率较低。
发明内容
为了解决现有技术的问题，本申请实施例提供了一种进行业务调度的方法和装置。所述技术方案如下：
第一方面，提供了一种进行业务调度的方法，所述方法包括：
当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况；
根据所述第一配置状况，通过训练得到的业务调度模型，生成替代资源列表，所述替代资源列表中记录有至少一个替代资源和所述至少一个替代资源对应的权值；
按照预设的第一挑选策略，根据所述至少一个替代资源对应的权值，在所述替代资源列表中选择第一替代资源；
基于所述第一替代资源，在所述CDN全网中进行业务调度。
可选的，所述方法还包括：
当在CDN测试环境中检测到预设的调度触发事件时，获取所述CDN测试环境中网络资源当前的第二配置状况，其中，所述CDN测试环境为所述CDN局部网络和/或CDN虚拟全网；
基于所述第二配置状况，对初始的业务调度模型进行训练，调整所述业务调度模型的参数。
可选的，所述基于所述第二配置状况，对初始的业务调度模型进行训练，调整所述业务调度模型的参数，包括：
基于所述第二配置状况，通过初始的业务调度模型，生成替代资源列表；
按照所述第一挑选策略，在所述替代资源列表中选择第二替代资源，并基于所述第二替代资源，在所述CDN测试环境中进行业务调度；
通过预设的网络评分标准，对所述业务调度后的，所述CDN测试环境中网络资源的第三配置状况进行评分；
根据评分结果，通过反向传播算法，对所述业务调度模型的参数进行调整。
可选的，所述在所述CDN全网中进行业务调度之后，还包括：
通过预设的网络评分标准，对所述业务调度后的，所述CDN全网中网络资源的第四配置状况进行评分；
根据评分结果，通过反向传播算法，对所述业务调度模型的参数进行调整。
可选的，所述当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况，包括：
当检测到所述CDN全网中第一网络资源的状态参数异常时，通过训练得到的故障报警模型，确定调度触发概率；
生成0到1之间的随机数，当所述随机数小于或等于所述调度触发概率时，获取CDN全网中网络资源当前的第一配置状况。
可选的，所述方法还包括：
根据所述CDN全网的故障报警历史数据，以马尔科夫链的形式计算建立故障报警模型。
第二方面，提供了一种进行业务调度的装置，所述装置包括：
获取模块，用于当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况；
生成模块，用于根据所述第一配置状况，通过训练得到的业务调度模型，生成替代资源列表，所述替代资源列表中记录有至少一个替代资源和所述至少一个替代资源对应的权值；
选择模块，用于按照预设的第一挑选策略，根据所述至少一个替代资源对应的权值，在所述替代资源列表中选择第一替代资源；
调度模块，用于基于所述第一替代资源，在所述CDN全网中进行业务调度。
可选的，所述获取模块，还用于：当在CDN测试环境中检测到预设的调度触发事件时，获取所述CDN测试环境中网络资源当前的第二配置状况，其中，所述CDN测试环境为所述CDN局部网络和/或CDN虚拟全网；
所述装置还包括训练模块，用于基于所述第二配置状况，对初始的业务调度模型进行训练，调整所述业务调度模型的参数。
可选的，所述训练模块，具体用于：
基于所述第二配置状况，通过初始的业务调度模型，生成替代资源列表；
按照所述第一挑选策略，在所述替代资源列表中选择第二替代资源，并基于所述第二替代资源，在所述CDN测试环境中进行业务调度；
通过预设的网络评分标准，对所述业务调度后的，所述CDN测试环境中网络资源的第三配置状况进行评分；
根据评分结果，通过反向传播算法，对所述业务调度模型的参数进行调整。
可选的，所述装置还包括：
评分模块，用于在所述CDN全网中进行业务调度之后，通过预设的网络评分标准，对所述业务调度后的，所述CDN全网中网络资源的第四配置状况进行评分；
调整模块，用于根据评分结果，通过反向传播算法，对所述业务调度模型的参数进行调整。
可选的，所述获取模块，具体用于：
当检测到所述CDN全网中第一网络资源的状态参数异常时，通过训练得到的故障报警模型，确定调度触发概率；
生成0到1之间的随机数，当所述随机数小于或等于所述调度触发概率时，获取CDN全网中网络资源当前的第一配置状况。
可选的，所述装置还包括：
建立模块，用于根据所述CDN全网的故障报警历史数据，以马尔科夫链的形式计算建立故障报警模型。
第三方面，提供了一种管理设备，所述管理设备包括处理器和存储器，所述存储器中存储有至少一条指令、至少一段程序、代码集或指令集，所述至少一条指令、所述至少一段程序、所述代码集或指令集由所述处理器加载并执行以实现如第一方面所述的进行业务调度的方法。
第四方面，提供了一种计算机可读存储介质，所述存储介质中存储有至少一条指令、至少一段程序、代码集或指令集，所述至少一条指令、所述至少一段程序、所述代码集或指令集由处理器加载并执行以实现如第一方面所述的进行业务调度的方法。
本申请实施例提供的技术方案带来的有益效果是：
本申请实施例中，当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况；根据第一配置状况，通过训练得到的业务调度模型，生成替代资源列表，替代资源列表中记录有至少一个替代资源和至少一个替代资源对应的权值；按照预设的第一挑选策略，根据至少一个替代资源对应的权值，在替代资源列表中选择第一替代资源；基于第一替代资源，在CDN全网中进行业务调度。这样，通过对业务调度过程以及业务调度结果反馈的不断学习，管理设备可以理解并掌握如何选择替代资源，避免人为指定调度规则所带来的不准确性、维护困难以及成本波动等问题，同时，在进行业务调度过程中，直接由管理设备通过业务调度模型选择出最优的替代资源，无需大量的人工参与，故而可以提高业务调度的效率。
附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本申请实施例提供的一种进行业务调度的方法流程图；
图2是本申请实施例提供的一种训练业务调度模型的方法流程图；
图3是本申请实施例提供的一种深度强化学习网络架构示意图；
图4是本申请实施例提供的一种进行业务调度的装置结构示意图；
图5是本申请实施例提供的一种进行业务调度的装置结构示意图；
图6是本申请实施例提供的一种进行业务调度的装置结构示意图；
图7是本申请实施例提供的一种进行业务调度的装置结构示意图；
图8是本申请实施例提供的一种管理设备的结构示意图。
具体实施方式
为使本申请的目的、技术方案和优点更加清楚，下面将结合附图对本申请实施例作进一步地详细描述。
本申请实施例提供了一种进行业务调度的方法，该方法的执行主体为CDN全网的管理设备，该管理设备用于监控CDN全网的网络资源的配置状况，并基于配置状况对CDN全网中的网络业务进行调度。其中，该管理设备可以是单独设立的，专用于管理的服务器，也可以是CDN全网中，同时支持网络业务的节点服务器。本实施例中管理设备的功能可以由独立的一台设备实现，也可以由多台设备组成的设备组共同实现。上述管理设备中可以包括处理器、存储器、收发器，处理器可以用于进行下述流程中的进行业务调度的处理，存储器可以用于存储下述处理过程中需要的数据以及产生的数据，收发器可以用于接收和发送下述处理过程中的相关数据。本实施例中，以管理设备为CDN全网的独立的管理设备为例进行说明，其它情况与之类似。
下面将结合具体实施例，对图1所示的处理流程进行详细的说明，内容可以如下：
步骤101，当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况。
其中，CDN全网为线上提供CDN加速服务的全部资源环境；网络资源可以是网络节点的处理资源和/或传输线路的传输资源，网络资源的配置状况可以包括各网络节点的CPU占用率、服务冗余度、输入输出值，以及各网络节点间的传输线路的带宽使用率、响应时间、丢包率等。
在实施中，CDN管理侧的技术人员可以预先在管理设备中设置进行业务 调度的调度触发事件，调度触发事件可以包括对CDN全网的故障预测报警、人为登记的故障发生、定时调度任务到达触发时间等。这样，管理设备可以通过实时监控上述调度触发事件是否发生，来决定是否进行业务调度处理。当检测到预设的调度触发事件时，可以先获取CDN全网中网络资源当前的第一配置状况。当然，在一定情况下，视业务调度规则的不同，可以在CDN全网中仅获取与业务调度相关联的部分网络资源的配置状况，例如，当某个网络节点持续超负荷运行，需要对该网络节点上的业务进行调度时，则可以仅获取该网络节点，及其附近的网络节点和传输线路上的网络资源的配置状况，后续涉及的获取的网络资源的配置状况的情况，具体处理与此处保持一致，不再一一赘述。
可选的，可以通过故障报警模型来提前触发业务调度，相应的，步骤101的处理可以如下：当检测到CDN全网中第一网络资源的状态参数异常时，通过训练得到的故障报警模型，确定调度触发概率；生成0到1之间的随机数，当随机数小于或等于调度触发概率时，获取CDN全网中网络资源当前的第一配置状况。
在实施中，管理设备中可以预先设置有用于预测CDN全网中故障的故障报警模型，通过该故障报警模型，可以通过CDN全网中的底层机器指标(即网络资源的状态参数)，来预测出现故障，需要进行业务调度的概率，若概率较高，则会以较大概率触发调度。具体的，管理设备可以获取CDN全网中网络资源的状态参数，当检测到CDN全网中第一网络资源的状态参数异常时，则可以通过训练得到的故障报警模型，确定调度触发概率。之后，可以生成0到1之间的随机数，然后将该随机数与上述调度触发概率作比较，当随机数小于或等于调度触发概率时，则触发业务调度，进而，管理设备可以获取CDN全网中网络资源当前的第一配置状况。上述处理可以具体由下式实现：
Figure PCTCN2018081867-appb-000001
其中，y＝1表示触发业务调度，y＝0表示不触发业务调度。P为通过故障报警模型确定的调度触发概率，π(t)为在时间t情况下，随机产生的0到1之间的随机数，如果该随机数小于或等于P，则y＝1；否则y＝0。
可选的，上述故障报警模型的训练过程可以如下：根据CDN全网的故障报警历史数据，以马尔科夫链的形式计算建立故障报警模型。
在实施中，管理设备可以根据CDN全网内近期的故障报警历史数据建立故障报警模型，并且通过CDN全网中实时产生的故障报警数据不断更新故障报警模型。其中，故障报警历史数据是对CDN全网中各个节点服务器监控产生的，从状态参数异常到实际触发业务调度的历史数据。管理设备可以基于故障报警历史数据，设定状态参数异常到触发业务调度的转换规则(也可称为底层机器指标异常报警到高层报警的转换规则)，比如在时间序列基础上，高层报警是基于底层指标异常报警产生的，且高层报警仅与其直接的下一级报警相关，故而可以采用马尔科夫链进行计算：
P(X n+1＝x|X 1＝x 1,X 2＝x 2,…,X n＝x n)＝P(X n+1＝x|X n＝x n)
其中，马尔科夫链是具有马尔科夫性质的随机变量的一个数列，其描述了一种状态序列，每个状态值取决于前面有限个状态。这些随机变量的范围，是它们所有可能取值的集合，被称为“状态空间”。具体到本实施例，随机变量X n代表在时间n时的状态，X n+1则仅是X n的一个函数。进一步的，管理设备可以根据CDN全网的故障报警历史数据，以马尔科夫链的形式计算建立故障报警模型。这样，通过对网络资源的状态参数进行分析，建立状态参数异常与触发业务调度间的关联关系，在检测状态参数变化的过程中，对后续故障进行预测，从而可以实现事前调度，一定程度上保障了服务质量和用户体验。
步骤102，根据第一配置状况，通过训练得到的业务调度模型，生成替代资源列表。
其中，替代资源列表中记录有至少一个替代资源和至少一个替代资源对应的权值。
在实施中，管理设备中可以预先存储有经过大量学习训练的业务调度模型，该业务调度模型可以是建立在大规模离散状态-行动空间上的深度强化学习模型，并通过样本训练进行迭代学习。传统的强化学习是一种解决离散状态-行动空间问题的算法，但是当离散状态和行动空间过大时，传统的强化学习则很难有效地处理这些情况。而深度强化学习模型将强化学习和深度学习两种机器学习方式结合起来，可以支持大规模离散状态-行动空间上的学习任务，适用于本实施例中进行业务调度的处理，具体的模型训练可以参见后文。
管理设备在获取CDN全网中网络资源当前的第一配置状况后，可以由第一配置状况得到当前全网资源配置的状态向量，然后调取训练得到的业务调度模型，并将上述状态向量作为业务调度模型的输入，采用业务调度模型的隐藏层的前向传播算法进行计算，进而可以由业务调度模型输出层得到替代资源列表(也可理解为替代资源概率列表)，该替代资源列表中记录有至少一个替代资源和至少一个替代资源对应的权值，此处，权值可以理解为对应的每个替代资源的选择权重。
步骤103，按照预设的第一挑选策略，根据至少一个替代资源对应的权值，在替代资源列表中选择第一替代资源。
在实施中，管理设备在通过业务调度模型得到替代资源列表后，可以按照预设的第一挑选策略，根据替代资源列表中记录的每个替代资源对应的权值，在上述替代资源列表中选择出第一替代资源，即选择出最优的替代资源。需要说明的是，第一挑选策略是对业务调度模型进行训练时所使用的挑选策略，此处并不对第一挑选策略为何种策略进行具体限定，管理侧的技术人员可以根据实际情况选择挑选策略，例如可以是贪心策略，即在替代资源列表中选择出对应的权值最高的替代资源作为第一替代资源，还可以先选择出权值最高的N个替代资源，然后再从中随机选出第一替代资源。
步骤104，基于第一替代资源，在CDN全网中进行业务调度。
在实施中，管理设备在替代资源列表中选择出第一替代资源后，可以基 于第一替代资源，在CDN全网中进行业务调度处理，即通过第一替代资源来执行相应业务。
可选的，管理设备在CDN全网中使用业务调度模型进行业务调度前，可以先在CDN测试环境中对业务调度模型进行参数调整，相应的处理可以如下：当在CDN测试环境中检测到预设的调度触发事件时，获取CDN测试环境中网络资源当前的第二配置状况；基于第二配置状况，对初始的业务调度模型进行训练，调整业务调度模型的参数。
其中，CDN测试环境为CDN局部网络和/或CDN虚拟全网。
在实施中，管理设备可以构建用于训练业务调度模型的CDN测试环境，在该CDN测试环境中，可以基于各种调度触发事件，对初始的业务调度模型进行迭代训练，以得到较为成熟的业务调度模型。具体的处理可以如下，当在CDN测试环境中检测到预设的调度触发事件时，管理设备可以获取CDN测试环境中网络资源当前的第二配置状况，然后基于第二配置状况，对初始的业务调度模型进行训练，以调整业务调度模型的参数。
上述CDN测试环境可以为CDN局部网络和/或CDN虚拟全网。其中，一方面，采用CDN局部网络的目的在于通过从CDN全网环境中规划出一个小范围的测试环境，以使业务调度模型在训练期间所产生的负面影响，可以被控制在局部网络的较小范围内，从而可以有效减小对CDN全网的服务质量和成本带来的影响。该局部网络可以根据网络节点的物理位置、业务所属运营商或者用户等级等进行划分。另一方面，CDN虚拟全网可以是指根据真实CDN全网虚拟出的一个网络环境，该网络环境支持业务调度系统与其进行交互，且支持根据调度结果，模拟真实CDN全网对调度结果的响应情况，此外，调度触发事件可以简单的由元胞自动机模型进行模拟。
不难发现，相对于CDN局部网络，基于CDN虚拟全网来训练业务调度模型存在以下几点优势：首先，CDN局部网络属于真实网络环境，在进行业务调度前，需要等待调度触发事件出现，才能进行调度处理以累积训练数据。 这样，模型训练需要一个比较长的时间周期。而对于CDN虚拟全网，可以通过元胞自动机进行调度触发事件的模拟，从而可以大幅缩短模型训练的时间周期。其次，CDN局部网络是为了减小业务调度模型在训练期间，对线上服务质量和成本带来较大的负面影响，而在CDN全网中规划出的一个子集环境，该子集环境虽然能够将负面影响限制在一个较小的范围内，但是却不能完全消除负面影响。同时，由于是CDN全网的子集环境，在该子集环境上产生的用于模型训练的数据是一个局部数据集，可能与CDN全网的全局数据集在数据分布存在很大差别，这样，会导致训练生成的业务调度模型出现过拟合现象，无法的泛化到CDN全网的环境中。而CDN虚拟网络中的任何处理都不会对真实的CDN全网造成影响，并且CDN虚拟网络是模拟真实的CDN全网环境，可以有效避免数据层面的局限性。
可选的，上述业务调度模型的训练处理可以具体包括如图2所示的几个步骤：
步骤201，基于第二配置状况，通过初始的业务调度模型，生成替代资源列表。
在实施中，管理设备在获取到CDN测试环境中网络资源当前的第二配置状况后，可以由第二配置状况得到当前全网资源配置的状态向量，再调取初始的业务调度模型，并将上述状态向量作为该初始的业务调度模型的输入，利用业务调度模型的隐藏层的前向传播算法进行计算，进而可以由业务调度模型的输出层得到替代资源列表，该替代资源列表中记录有至少一个替代资源和至少一个替代资源对应的权值。
业务调度模型的结构可以参考图3所示的深度强化学习网络架构，其中，输入层为状态向量，输出层为替代资源的权值，隐藏层为神经元线性或非线性计算。业务调度模型可以基于初始化的参数，根据输入层输入的状态向量，经过隐藏层前向传播计算，从而输出层输出权值，相应的计算公式可以如下：
Figure PCTCN2018081867-appb-000002
Figure PCTCN2018081867-appb-000003
其中，
Figure PCTCN2018081867-appb-000004
表示第l+1层第i个单元输入加权和，x 1，x 2，…，x n为输入单元，
Figure PCTCN2018081867-appb-000005
为第l层第i个单元的偏置单元，
Figure PCTCN2018081867-appb-000006
为第l层的第j个单元到第l+1层的第i个单元之间的连接权重。
Figure PCTCN2018081867-appb-000007
表示第l+1层第i个单元的激活值(即相当于上述替代资源对应的权值)，f(·)为激活函数，可以是sigmoid、tanh等函数。
步骤202，按照第一挑选策略，在替代资源列表中选择第二替代资源，并基于第二替代资源，在CDN测试环境中进行业务调度。
在实施中，管理设备在通过业务调度模型得到替代资源列表后，可以按照第一挑选策略，根据替代资源列表中记录的每个替代资源对应的权值，在上述替代资源列表中选择出第二替代资源。然后，管理设备可以在CDN测试环境中，基于第二替代资源进行业务调度。值得一提的是，属于业务调度模型的训练阶段，上述替代资源列表中各替代资源的权值准确性无法得到保证，故而选择出的第二替代资源有可能不是实际最优的替代资源。
步骤203，通过预设的网络评分标准，对业务调度后的，CDN测试环境中网络资源的第三配置状况进行评分。
在实施中，管理侧的技术人员可以针对网络节点的带宽情况、服务能力、用户覆盖情况、响应时间等多维度数据进行综合考虑，建立网络质量评分标准，同时还可以对网络节点的规划情况、计费类型、计费系数等多维度数据进行综合考虑，建立网络成本评分标准。具体的，首先可以从网络资源角度进行质量评分，如对单个节点服务器的质量评分、服务集群质量评分等，还可以从用户角度进行质量评分，比如用户响应时间等，然后对两个维度的质量分数进行综合分析；然后还可以基于计费单元(比如单个网络节点)，根据不同的计费类型和计费系数计算出当前CDN所产生的成本分数，最后可以将质量分数和成本分数综合起来，生成网络评分标准，并将该网络评分标准存储至管理设备中。
这样，在基于第二替代资源进行业务调度之后，管理设备可以通过上述预设的网络评分标准，对调度后的CDN测试环境中网络资源的第三配置状况进行评分。
步骤204，根据评分结果，通过反向传播算法，对业务调度模型的参数进行调整。
在实施中，管理设备获取到调度后的CDN测试环境中网络资源的配置状况的评分结果后，可以通过反向传播算法，对业务调度模型的参数进行调整。其中，反向传播算法可以如下：
公式1，计算业务调度模型的最后一层计算所产生的误差：
Figure PCTCN2018081867-appb-000008
公式2，计算业务调度模型的每一层计算所产生的误差：
Figure PCTCN2018081867-appb-000009
公式3，计算权重的梯度：
Figure PCTCN2018081867-appb-000010
公式4，计算偏置的梯度：
Figure PCTCN2018081867-appb-000011
其中，L表示隐藏层的最后一层，C表示模型输出值和标准值之间的真实误差，δ L表示计算得到的最后一层的误差，δ l表示计算得到的第l层的误差，⊙为Hadamard(阿达玛)乘积，
Figure PCTCN2018081867-appb-000012
表示最后一层误差对最后一层单元输出a的导数，σ′(z L)表示激活函数对最后一层单元输入的导数，σ′(z l)表示激活函数对第l层单元输入的导数。
可选的，与训练阶段相似，步骤104在CDN全网中进行业务调度后，可以对调度结果进行评分，然后基于评分结果对参数调整，相应的处理可以如下：通过预设的网络评分标准，对业务调度后的，CDN全网中网络资源的第四配置状况进行评分；根据评分结果，通过反向传播算法，对业务调度模型的参数进行调整。
在实施中，管理设备基于第一替代资源，在CDN全网中进行业务调度 之后，可以获取业务调度后的CDN全网中网络资源的第四配置状况，然后可以通过预设的网络评分标准，对第四配置状况进行评分。之后，可以再根据评分结果，通过反向传播算法，对业务调度模型的参数进行调整。
这样，业务调度模型利用了深度神经网络的高度抽象能力，可以有效地对CDN全网环境中的复杂问题进行表征，并根据网络评分标准的反馈，通过反向传播算法更新模型中的参数，从而在经历一段时间的学习过程后，能够对业务调度问题进行综合考量，选择出全局最优的方案。在训练阶段，若出现触发调度的情况，将CDN测试环境中网络资源的配置状况设置为模型的输入，模型会基于当前的参数，输出替代资源列表，之后再根据一定挑选策略选择替代资源，然后利用网络评分标准对替代资源的服务情况进行评估并反馈给模型，模型接收到反馈后，根据反向传播算法，调整模型参数，使后续调度能够朝着全局最优的方向进行。当训练达到一定迭代次数或者训练结果满足某个预定条件时，可以生成性能较好的业务调度模型。之后，可以将该业务调度模型应用到CDN全网环境中，并且在后续的运行过程中仍然持续进行学习，即发生调度触发事件时，业务调度模型可以选择预计对未来产生正面影响的替代资源继续执行业务，同时可以通过网络评分标准对替代资源的选择进行评估，从而业务调度模型可以根据评分结果进一步的调整参数。
本申请实施例中，当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况；根据第一配置状况，通过训练得到的业务调度模型，生成替代资源列表，替代资源列表中记录有至少一个替代资源和至少一个替代资源对应的权值；按照预设的第一挑选策略，根据至少一个替代资源对应的权值，在替代资源列表中选择第一替代资源；基于第一替代资源，在CDN全网中进行业务调度。这样，通过对业务调度过程以及业务调度结果反馈的不断学习，理解并掌握应该如何选择替代资源，避免人为指定调度规则带来的不准确性、维护困难以及成本波动等问题，同时，在进行业务调度过程中， 直接由管理设备通过业务调度模型选择出最优的替代资源，无需大量的人工参与，故而可以提高业务调度的效率。
基于相同的技术构思，本申请实施例还提供了一种进行业务调度的装置，如图4所示，该装置包括：
获取模块401，用于当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况；
生成模块402，用于根据所述第一配置状况，通过训练得到的业务调度模型，生成替代资源列表，所述替代资源列表中记录有至少一个替代资源和所述至少一个替代资源对应的权值；
选择模块403，用于按照预设的第一挑选策略，根据所述至少一个替代资源对应的权值，在所述替代资源列表中选择第一替代资源；
调度模块404，用于基于所述第一替代资源，在所述CDN全网中进行业务调度。
可选的，所述获取模块401，还用于：当在CDN测试环境中检测到预设的调度触发事件时，获取所述CDN测试环境中网络资源当前的第二配置状况，其中，所述CDN测试环境为所述CDN局部网络和/或CDN虚拟全网；
如图5所示，所述装置还包括训练模块405，用于基于所述第二配置状况，对初始的业务调度模型进行训练，调整所述业务调度模型的参数。
可选的，所述训练模块405，具体用于：
基于所述第二配置状况，通过初始的业务调度模型，生成替代资源列表；
按照所述第一挑选策略，在所述替代资源列表中选择第二替代资源，并基于所述第二替代资源，在所述CDN测试环境中进行业务调度；
通过预设的网络评分标准，对所述业务调度后的，所述CDN测试环境中网络资源的第三配置状况进行评分；
根据评分结果，通过反向传播算法，对所述业务调度模型的参数进行调整。
可选的，如图6所示，所述装置还包括：
评分模块406，用于在所述CDN全网中进行业务调度之后，通过预设的网络评分标准，对所述业务调度后的，所述CDN全网中网络资源的第四配置状况进行评分；
调整模块407，用于根据评分结果，通过反向传播算法，对所述业务调度模型的参数进行调整。
可选的，所述获取模块401，具体用于：
当检测到所述CDN全网中第一网络资源的状态参数异常时，通过训练得到的故障报警模型，确定调度触发概率；
生成0到1之间的随机数，当所述随机数小于或等于所述调度触发概率时，获取CDN全网中网络资源当前的第一配置状况。
可选的，如图7所示，所述装置还包括：
建立模块408，用于根据所述CDN全网的故障报警历史数据，以马尔科夫链的形式计算建立故障报警模型。
本申请实施例中，当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况；根据第一配置状况，通过训练得到的业务调度模型，生成替代资源列表，替代资源列表中记录有至少一个替代资源和至少一个替代资源对应的权值；按照预设的第一挑选策略，根据至少一个替代资源对应的权值，在替代资源列表中选择第一替代资源；基于第一替代资源，在CDN全网中进行业务调度。这样，通过对业务调度过程以及业务调度结果反馈的不断学习，理解并掌握应该如何选择替代资源，避免人为指定调度规则带来的不准确性、维护困难以及成本波动等问题，同时，在进行业务调度过程中，直接由管理设备通过业务调度模型选择出最优的替代资源，无需大量的人工参与，故而可以提高业务调度的效率。
需要说明的是：上述实施例提供的进行业务调度的装置在进行业务调度时，仅以上述各功能模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能模块完成，即将装置的内部结构划分成不同的功能模块，以完成以上描述的全部或者部分功能。另外，上述实施例提供的进行业务调度的装置与进行业务调度的方法实施例属于同一构思，其具体实现过程详见方法实施例，这里不再赘述。
图8是本申请实施例提供的管理设备的结构示意图。该管理设备800可因配置或性能不同而产生比较大的差异，可以包括一个或一个以上中央处理器822(例如，一个或一个以上处理器)和存储器832，一个或一个以上存储应用程序842或数据844的存储介质830(例如一个或一个以上海量存储设备)。其中，存储器832和存储介质830可以是短暂存储或持久存储。存储在存储介质830的程序可以包括一个或一个以上模块(图示没标出)，每个模块可以包括对服务端中的一系列指令操作。更进一步地，中央处理器822可以设置为与存储介质830通信，在管理设备800上执行存储介质830中的一系列指令操作。
管理设备800还可以包括一个或一个以上电源826，一个或一个以上有线或无线网络接口850，一个或一个以上输入输出接口858，一个或一个以上键盘856，和/或，一个或一个以上操作系统841，例如WindowsServerTM，MacOSXTM，UnixTM，LinuxTM，FreeBSDTM等等。
管理设备800可以包括有存储器，以及一个或者一个以上的程序，其中一个或者一个以上程序存储于存储器中，且经配置以由一个或者一个以上处理器执行所述一个或者一个以上程序包含用于执行上述进行业务调度的指令。
本领域普通技术人员可以理解实现上述实施例的全部或部分步骤可以通过硬件来完成，也可以通过程序来指令相关的硬件完成，所述的程序可以存 储于一种计算机可读存储介质中，上述提到的存储介质可以是只读存储器，磁盘或光盘等。
以上所述仅为本申请的较佳实施例，并不用以限制本申请，凡在本申请的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本申请的保护范围之内。

Claims (13)

  1. 一种进行业务调度的方法，其特征在于，所述方法包括：
    当检测到预设的调度触发事件时，获取内容分发网络CDN全网中网络资源当前的第一配置状况；
    根据所述第一配置状况，通过训练得到的业务调度模型，生成替代资源列表，所述替代资源列表中记录有至少一个替代资源和所述至少一个替代资源对应的权值；
    按照预设的第一挑选策略，根据所述至少一个替代资源对应的权值，在所述替代资源列表中选择第一替代资源；
    基于所述第一替代资源，在所述CDN全网中进行业务调度。
  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：
    当在CDN测试环境中检测到预设的调度触发事件时，获取所述CDN测试环境中网络资源当前的第二配置状况，其中，所述CDN测试环境为所述CDN局部网络和/或CDN虚拟全网；
    基于所述第二配置状况，对初始的业务调度模型进行训练，调整所述业务调度模型的参数。
  3. 根据权利要求2所述的方法，其特征在于，所述基于所述第二配置状况，对初始的业务调度模型进行训练，调整所述业务调度模型的参数，包括：
    基于所述第二配置状况，通过初始的业务调度模型，生成替代资源列表；按照所述第一挑选策略，在所述替代资源列表中选择第二替代资源，并基于所述第二替代资源，在所述CDN测试环境中进行业务调度；通过预设的网络评分标准，对所述业务调度后的，所述CDN测试环境中网络资源的第三配置状况进行评分；
    根据评分结果，通过反向传播算法，对所述业务调度模型的参数进行调整。
  4. 根据权利要求1所述的方法，其特征在于，在所述CDN全网中进行业务调度之后，还包括：
    通过预设的网络评分标准，对所述业务调度后的，所述CDN全网中网络资源的第四配置状况进行评分；
    根据评分结果，通过反向传播算法，对所述业务调度模型的参数进行调整。
  5. 根据权利要求1至4任一所述的方法，其特征在于，所述当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况，包括：
    当检测到所述CDN全网中第一网络资源的状态参数异常时，通过训练得到的故障报警模型，确定调度触发概率；
    生成0到1之间的随机数，当所述随机数小于或等于所述调度触发概率时，获取所述CDN全网中网络资源当前的第一配置状况。
  6. 根据权利要求5所述的方法，其特征在于，所述方法还包括：
    根据所述CDN全网的故障报警历史数据，以马尔科夫链的形式计算建立所述故障报警模型。
  7. 一种进行业务调度的装置，其特征在于，所述装置包括：
    获取模块，用于当检测到预设的调度触发事件时，获取CDN全网中网络资源当前的第一配置状况；
    生成模块，用于根据所述第一配置状况，通过训练得到的业务调度模型，生成替代资源列表，所述替代资源列表中记录有至少一个替代资源和所述至少一个替代资源对应的权值；
    选择模块，用于按照预设的第一挑选策略，根据所述至少一个替代资源对 应的权值，在所述替代资源列表中选择第一替代资源；
    调度模块，用于基于所述第一替代资源，在所述CDN全网中进行业务调度。
  8. 根据权利要求7所述的装置，其特征在于，所述获取模块，还用于：当在CDN测试环境中检测到预设的调度触发事件时，获取所述CDN测试环境中网络资源当前的第二配置状况，其中，所述CDN测试环境为所述CDN局部网络和/或CDN虚拟全网；
    所述装置还包括训练模块，用于基于所述第二配置状况，对初始的业务调度模型进行训练，调整所述业务调度模型的参数。
  9. 根据权利要求8所述的装置，其特征在于，所述训练模块，具体用于：
    基于所述第二配置状况，通过初始的业务调度模型，生成替代资源列表；
    按照所述第一挑选策略，在所述替代资源列表中选择第二替代资源，并基于所述第二替代资源，在所述CDN测试环境中进行业务调度；
    通过预设的网络评分标准，对所述业务调度后的，所述CDN测试环境中网络资源的第三配置状况进行评分；
    根据评分结果，通过反向传播算法，对所述业务调度模型的参数进行调整。
  10. 根据权利要求7所述的装置，其特征在于，所述装置还包括：
    评分模块，用于在所述CDN全网中进行业务调度之后，通过预设的网络评分标准，对所述业务调度后的，所述CDN全网中网络资源的第四配置状况进行评分；
    调整模块，用于根据评分结果，通过反向传播算法，对所述业务调度模型的参数进行调整。
  11. 根据权利要求7-10任一所述的装置，其特征在于，所述获取模块，具体用于：
    当检测到所述CDN全网中第一网络资源的状态参数异常时，通过训练得到的故障报警模型，确定调度触发概率；
    生成0到1之间的随机数，当所述随机数小于或等于所述调度触发概率时，获取所述CDN全网中网络资源当前的第一配置状况。
  12. 根据权利要求11所述的装置，其特征在于，所述装置还包括：
    建立模块，用于根据所述CDN全网的故障报警历史数据，以马尔科夫链的形式计算建立所述故障报警模型。
  13. 一种管理设备，其特征在于，所述管理设备包括处理器和存储器，所述存储器中存储有至少一条指令、至少一段程序、代码集或指令集，所述至少一条指令、所述至少一段程序、所述代码集或指令集由所述处理器加载并执行以实现如权利要求1至6任一所述的进行业务调度的方法。
PCT/CN2018/081867 2017-07-14 2018-04-04 一种进行业务调度的方法和装置 WO2019011015A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
EP18832346.3A EP3629553B1 (en) 2017-07-14 2018-04-04 Method and device for service scheduling
US16/726,115 US11128684B2 (en) 2017-07-14 2019-12-23 Method and apparatus for scheduling service

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710576106.X 2017-07-14
CN201710576106.XA CN107231436B (zh) 2017-07-14 2017-07-14 一种进行业务调度的方法和装置

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/726,115 Continuation US11128684B2 (en) 2017-07-14 2019-12-23 Method and apparatus for scheduling service

Publications (1)

Publication Number Publication Date
WO2019011015A1 true WO2019011015A1 (zh) 2019-01-17

Family

ID=59956825

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/081867 WO2019011015A1 (zh) 2017-07-14 2018-04-04 一种进行业务调度的方法和装置

Country Status (4)

Country Link
US (1) US11128684B2 (zh)
EP (1) EP3629553B1 (zh)
CN (1) CN107231436B (zh)
WO (1) WO2019011015A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112243025A (zh) * 2020-09-22 2021-01-19 网宿科技股份有限公司 节点成本的调度方法、电子设备及存储介质
CN114629906A (zh) * 2022-03-14 2022-06-14 浙江大学 一种可靠的基于深度强化学习的云容器集群资源调度方法及装置
CN114726922A (zh) * 2022-03-31 2022-07-08 联想(北京)有限公司 网络资源调度方法、装置及电子设备和存储介质

Families Citing this family (24)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107231436B (zh) 2017-07-14 2021-02-02 网宿科技股份有限公司 一种进行业务调度的方法和装置
CN108306769B (zh) * 2018-01-31 2021-08-20 北京奇艺世纪科技有限公司 一种cdn的节点部署控制方法和系统
EP3528435B1 (en) * 2018-02-16 2021-03-31 Juniper Networks, Inc. Automated configuration and data collection during modeling of network devices
CN110276503B (zh) * 2018-03-14 2023-04-21 吉旗物联科技（上海）有限公司 一种自动识别冷链车辆任务的方法
CN109102515A (zh) * 2018-07-31 2018-12-28 浙江杭钢健康产业投资管理有限公司 一种基于多列深浅卷积神经网络的细胞计数方法
CN109388484B (zh) * 2018-08-16 2020-07-28 广东石油化工学院 一种基于Deep Q-network算法的多资源云作业调度方法
CN110519365B (zh) * 2019-08-26 2022-07-01 网宿科技股份有限公司 一种变更设备业务的方法和业务变更系统
CN113132437B (zh) * 2019-12-31 2024-01-23 中兴通讯股份有限公司 Cdn调度方法、系统、设备和存储介质
CN113067714B (zh) * 2020-01-02 2022-12-13 中国移动通信有限公司研究院 一种内容分发网络调度处理方法、装置及设备
CN113391888B (zh) * 2020-03-12 2024-05-17 北京京东振世信息技术有限公司 业务资源的调度方法、装置、设备及存储介质
CN111614484B (zh) * 2020-04-13 2021-11-02 网宿科技股份有限公司 一种节点流量的调入、恢复方法、系统及中心服务器
CN111652635A (zh) * 2020-05-14 2020-09-11 拉扎斯网络科技（上海）有限公司 资源策略信息推荐、执行状态数据、预测模型获得方法
CN111589109B (zh) * 2020-05-21 2023-09-26 网易（杭州）网络有限公司 一种数据处理方法、装置、电子设备及存储介质
CN112016789B (zh) * 2020-07-15 2024-01-16 北京淇瑀信息科技有限公司 互联网金融业务处理方法、装置和电子设备
CN111950872A (zh) * 2020-07-30 2020-11-17 深圳市高德信通信股份有限公司 一种cdn资源调度系统
CN111953538A (zh) * 2020-07-31 2020-11-17 深圳市高德信通信股份有限公司 一种基于大数据处理的cdn带宽调度系统
CN112667398B (zh) * 2020-12-28 2023-09-01 北京奇艺世纪科技有限公司 资源调度方法、装置、电子设备及存储介质
CN114978915B (zh) * 2021-02-24 2024-01-19 北京金山云网络技术有限公司 Cdn节点带宽规划方法、装置、电子设备及存储介质
CN115379243B (zh) * 2021-05-21 2024-02-27 北京字跳网络技术有限公司 Cdn的调度方法、装置、设备及存储介质
CN113839813B (zh) * 2021-09-14 2023-03-24 中国电信股份有限公司 设备控制策略模型训练及设备控制方法、介质与设备
CN114006827B (zh) * 2021-09-27 2023-01-24 清华大学 时间触发调度模型的训练、时间触发调度方法及装置
CN114244903B (zh) * 2021-11-01 2024-05-28 网宿科技股份有限公司 资源调度方法、系统、服务器及存储介质
CN114389942B (zh) * 2022-01-11 2024-06-04 北京金山云网络技术有限公司 Cdn配置方法和装置、电子设备和存储介质
CN115426319B (zh) * 2022-08-30 2024-09-06 商飞智能技术有限公司 一种网络资源调度系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101694652A (zh) * 2009-09-30 2010-04-14 西安交通大学 一种基于极速神经网络的网络资源个性化推荐方法
CN103051546A (zh) * 2012-12-12 2013-04-17 中国科学院计算技术研究所 一种基于迟滞调度的网络流量冲突避免方法及系统
CN103167028A (zh) * 2013-02-21 2013-06-19 胡凤兵 节省传输带宽并能增大并发量的用户收敛方法
WO2013189024A1 (en) * 2012-06-19 2013-12-27 Hewlett-Packard Development Company, L.P. Server site selection
CN105956661A (zh) * 2016-04-15 2016-09-21 中山大学 一种在sdn网络上实现dann在线训练的系统
CN106940801A (zh) * 2016-01-04 2017-07-11 中国科学院声学研究所 一种用于广域网络的深度强化学习推荐系统及方法
CN107231436A (zh) * 2017-07-14 2017-10-03 网宿科技股份有限公司 一种进行业务调度的方法和装置

Family Cites Families (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9549043B1 (en) * 2004-07-20 2017-01-17 Conviva Inc. Allocating resources in a content delivery environment
US8214450B2 (en) * 2005-08-01 2012-07-03 Limelight Networks, Inc. Dynamic bandwidth allocation
US8103282B2 (en) * 2005-09-28 2012-01-24 Avaya Inc. Methods and apparatus for allocating resources in a distributed environment based on network assessment
US20080215409A1 (en) * 2007-01-03 2008-09-04 Victorware, Llc Iterative resource scheduling
US20090307215A1 (en) * 2008-06-04 2009-12-10 Tynt Multimedia, Inc. Network resource annotation and search system
US9002978B2 (en) * 2011-07-08 2015-04-07 Ming Li Content delivery prediction and feedback systems
US9438487B2 (en) * 2012-02-23 2016-09-06 Ericsson Ab Bandwith policy management in a self-corrected content delivery network
US9100464B2 (en) * 2012-08-29 2015-08-04 Ericsson Television Inc. Regulating content streams from a weighted fair queuing scheduler using weights defined for user equipment nodes
CN103117876B (zh) * 2013-01-24 2018-10-26 中兴通讯股份有限公司 用户状态信息的同步方法及装置
CN103412875B (zh) * 2013-07-11 2016-09-28 广东工业大学 基于ahp决策模型的cdn策略调整方法
US20160283304A1 (en) * 2013-12-20 2016-09-29 Hitachi, Ltd. Performance prediction method, performance prediction system and program
TWI603266B (zh) * 2014-03-03 2017-10-21 廣達電腦股份有限公司 虛擬機器之資源調整方法及系統
WO2015179778A1 (en) * 2014-05-23 2015-11-26 Datarobot Systems and techniques for predictive data analytics
CN104320487B (zh) * 2014-11-11 2018-03-20 网宿科技股份有限公司 内容分发网络的http调度系统和方法
CN104834722B (zh) * 2015-05-12 2018-03-02 网宿科技股份有限公司 基于cdn的内容管理系统
CN106713265B (zh) * 2016-11-21 2019-05-28 清华大学深圳研究生院 Cdn节点分配方法及装置、cdn节点分配服务器及cdn网络系统

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101694652A (zh) * 2009-09-30 2010-04-14 西安交通大学 一种基于极速神经网络的网络资源个性化推荐方法
WO2013189024A1 (en) * 2012-06-19 2013-12-27 Hewlett-Packard Development Company, L.P. Server site selection
CN103051546A (zh) * 2012-12-12 2013-04-17 中国科学院计算技术研究所 一种基于迟滞调度的网络流量冲突避免方法及系统
CN103167028A (zh) * 2013-02-21 2013-06-19 胡凤兵 节省传输带宽并能增大并发量的用户收敛方法
CN106940801A (zh) * 2016-01-04 2017-07-11 中国科学院声学研究所 一种用于广域网络的深度强化学习推荐系统及方法
CN105956661A (zh) * 2016-04-15 2016-09-21 中山大学 一种在sdn网络上实现dann在线训练的系统
CN107231436A (zh) * 2017-07-14 2017-10-03 网宿科技股份有限公司 一种进行业务调度的方法和装置

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP3629553A4 *

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112243025A (zh) * 2020-09-22 2021-01-19 网宿科技股份有限公司 节点成本的调度方法、电子设备及存储介质
CN112243025B (zh) * 2020-09-22 2023-10-17 网宿科技股份有限公司 节点成本的调度方法、电子设备及存储介质
CN114629906A (zh) * 2022-03-14 2022-06-14 浙江大学 一种可靠的基于深度强化学习的云容器集群资源调度方法及装置
CN114629906B (zh) * 2022-03-14 2023-09-29 浙江大学 一种可靠的基于深度强化学习的云容器集群资源调度方法及装置
CN114726922A (zh) * 2022-03-31 2022-07-08 联想(北京)有限公司 网络资源调度方法、装置及电子设备和存储介质
CN114726922B (zh) * 2022-03-31 2023-11-21 联想(北京)有限公司 网络资源调度方法、装置及电子设备和存储介质

Also Published As

Publication number Publication date
EP3629553B1 (en) 2022-01-26
CN107231436B (zh) 2021-02-02
CN107231436A (zh) 2017-10-03
US20200137131A1 (en) 2020-04-30
EP3629553A1 (en) 2020-04-01
US11128684B2 (en) 2021-09-21
EP3629553A4 (en) 2020-07-08

Similar Documents

Publication Publication Date Title
WO2019011015A1 (zh) 一种进行业务调度的方法和装置
US11757982B2 (en) Performing load balancing self adjustment within an application environment
US11334831B2 (en) Predictive risk assessment in system modeling
CN112685170B (zh) 备份策略的动态优化
JP2019519871A (ja) 予測を使用してターゲットシステムを制御すること
US11283863B1 (en) Data center management using digital twins
Zyrianoff et al. Scalability of real-time iot-based applications for smart cities
Moradi et al. Performance prediction in dynamic clouds using transfer learning
Alipour et al. Online machine learning for cloud resource provisioning of microservice backend systems
Lockhart et al. Scission: Performance-driven and context-aware cloud-edge distribution of deep neural networks
CN113485801B (zh) 基于神经网络相似度建模的实时dnn调度系统及方法
Paul Martin et al. CREW: Cost and Reliability aware Eagle‐Whale optimiser for service placement in Fog
Dube et al. AI gauge: Runtime estimation for deep learning in the cloud
Wang et al. GeoCol: A geo-distributed cloud storage system with low cost and latency using reinforcement learning
CN117573459A (zh) 一种故障检测方法、装置、电子设备及存储介质
Jassas et al. A failure prediction model for large scale cloud applications using deep learning
AU2021218217A1 (en) Systems and methods for preventative monitoring using AI learning of outcomes and responses from previous experience.
KR20160044623A (ko) 리눅스 가상 서버의 로드 밸런싱 방법
Kordelas et al. KORDI: A Framework for Real-Time Performance and Cost Optimization of Apache Spark Streaming
Abdel Raouf et al. A predictive replication for multi‐tenant databases using deep learning
Bolodurina et al. Development and research of model for ensuring reliability of operation of network infrastructure objects in cyber-physical system located in the cloud platform
Wang et al. DeepScaling: Autoscaling Microservices With Stable CPU Utilization for Large Scale Production Cloud Systems
Xiong et al. Improving NoSQL's Performance Metrics via Machine Learning
Siriwardhana et al. A network science-based approach for an optimal microservice governance
Rattanaopas et al. Adaptive workload prediction for cloud-based server infrastructures

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: 18832346

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

ENP Entry into the national phase

Ref document number: 2018832346

Country of ref document: EP

Effective date: 20191223