Task scheduling method, device and system for collaborative edge computing

Abstract

The invention provides a task scheduling method, device and system for cooperative edge computing. The method comprises the following steps: receiving scheduling cost and overhead accumulation, task arrival prediction information, resource allocation information and network state information corresponding to each service reported by each network access point in a multi-region edge internet; the accumulation of the scheduling cost overhead is accumulation of the scheduling cost overhead of the network access point for scheduling task requests in a scheduling period for a service; at the end of the current scheduling cycle, determining a scheduling strategy of the next scheduling cycle based on task arrival prediction information, resource allocation information, network state information and scheduling cost overhead accumulation; the scheduling strategy is a scheduling probability of scheduling a task request of a service to an edge computing resource pool of each region; and before the next scheduling period starts, issuing the scheduling strategy determined for each service of each network access point to the corresponding network access point. According to the method, efficient edge collaborative calculation can be realized on the premise of controlling the scheduling task cost.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a task scheduling method, device and system for collaborative edge computing. Background Art

[0002] With the rapid development of IoT and 5G technologies, the demand for latency-sensitive tasks is increasing, such as face recognition, Internet of Vehicles, AR / VR, etc. Resource-constrained terminal devices cannot process these tasks locally, and traditional cloud computing introduces long backhaul link latency and cannot provide real-time computing. Mobile edge computing has become a promising paradigm that can assist various 5G service scenarios, including low-latency services.

[0003] However, compared with the cloud, edge computing nodes have limited resources. When faced with a large number of service requests, edge nodes are overloaded and cannot guarantee the quality of user service experience. Therefore, a cloud-edge collaboration solution is proposed, in which high-priority, latency-sensitive tasks are left to the edge for processing, while low-priority, latency-insensitive tasks are scheduled to the cloud for processing, alleviating the pressure on edge computing nodes.

[0004] However, when the main service requested is latency-sensitive, such as in factories, cloud-edge collaboration still cannot effectively alleviate edge pressure. Therefore, collaborative edge computing becomes a promising solution, in which edge computing nodes can collaborate with each other to process tasks, effectively coping with the spatiotemporal non-uniformity of user requests, while enhancing the computing power of the edge and effectively improving the user's service experience quality.

[0005] For example, among multiple factories operated by an enterprise, an edge node is deployed in each factory for privacy and latency requirements. The edge nodes between different factories have a natural cooperative relationship. That is, when the service requests of a factory surge, the edge node of the factory can dispatch some requests to the edge nodes of other factories for processing. This will not cause privacy leakage and enhance the performance of edge computing.

[0006] However, task scheduling between edge nodes often incurs additional cost overhead, mainly network cost. Obviously, cost is a key factor that needs to be considered, that is, efficient collaborative edge computing needs to be achieved under cost control to enhance the user's service experience quality. Summary of the invention

[0007] In view of this, the present application provides a task scheduling method, device and system for collaborative edge computing, which can achieve efficient collaborative edge computing while controlling the cost of scheduling tasks.

[0008] To solve the above technical problems, the technical solution of this application is implemented as follows:

[0009] In one embodiment, a task scheduling method for collaborative edge computing is provided, which is applied to a central controller in a multi-region edge interconnection network, and the method includes:

[0010] Receiving the scheduling cost accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service reported by each network access point in the multi-region edge interconnection network; the scheduling cost accumulation is the accumulation of scheduling cost of scheduling task requests of a service by the network access point in a scheduling cycle;

[0011] At the end of the current scheduling cycle, for each service of each network access point, a computing power network graph is calculated according to the corresponding task arrival prediction information, resource allocation information and network status information; and a scheduling strategy for the next scheduling cycle is determined according to the scheduling cost overhead accumulation and the computing power network graph; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network;

[0012] Before the next scheduling period starts, the scheduling strategy determined for each service of each network access point is sent to the corresponding network access point.

[0013] Wherein, determining the scheduling strategy for the next scheduling cycle according to the scheduling cost accumulation and the computing power network diagram includes:

[0014] Calculate the excess accumulation of scheduling cost for each service at the beginning of the next scheduling cycle according to the scheduling cost overhead accumulation corresponding to each service reported by each network access node and the preset long-term average scheduling cost budget;

[0015] The scheduling strategy for the next scheduling cycle is determined according to the excess accumulation of scheduling cost and the computing power network diagram.

[0016] Wherein, for any service, determining the scheduling strategy for the next scheduling cycle according to the excess accumulation of scheduling cost and the computing power network diagram comprises the steps of:

[0017] A. Randomly generate an initial region category allocation solution group; wherein the region categories include: source node, sink node and isolated node; wherein the source node has only out-degree but no in-degree; the sink node has only in-degree but no out-degree; the isolated node has neither out-degree nor in-degree;

[0018] B. For each area category allocation solution, solving the optimal scheduling strategy and the corresponding objective function value under the current area category allocation solution; the objective function value is determined by a Lyapunov shift plus penalty function, and the Lyapunov shift plus penalty function incorporates excess accumulation of scheduling costs into delay penalties;

[0019] C. According to each regional category allocation solution and the corresponding optimal scheduling strategy and objective function value, the fitness corresponding to the regional category allocation solution is obtained, and the regional category allocation solutions corresponding to the fitness values ​​greater than the preset fitness values ​​are selected to form an optimized regional category allocation solution group;

[0020] D. Determine whether the maximum number of iterations or the preset optimization condition is reached. If yes, execute step G; otherwise, execute step E;

[0021] E. Based on the optimized area category allocation solution group, a first extended area category allocation group is generated using a crossover probability and a crossover method;

[0022] F. Based on the first extended area category allocation solution group, generate a second extended area category allocation solution group using mutation probability and mutation method; execute step B;

[0023] G. Output an optimal scheduling strategy as the scheduling strategy determined according to the scheduling cost accumulation and the computing power network diagram; the optimal scheduling strategy is the strategy corresponding to the regional category allocation solution with the highest fitness in the optimized regional category allocation solution group.

[0024] Wherein, when solving the optimal scheduling strategy under the current area category allocation, it is obtained through a harmony search algorithm.

[0025] In one embodiment, a task scheduling method for collaborative edge computing is provided, which is applied to a network access point in any area of ​​a multi-area edge interconnection network, and the method includes:

[0026] In response to receiving the task request, determining a service corresponding to the task request;

[0027] Obtaining a scheduling strategy corresponding to the service in the current scheduling cycle; the scheduling strategy is a scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network; the scheduling probability is determined and issued by the central controller before the start of each scheduling cycle based on the scheduling cost overhead accumulation and the computing power network diagram of each service; the computing power network diagram is calculated based on the task arrival prediction information, resource allocation information and network status information reported by each network access point in the multi-area edge Internet; the scheduling cost overhead accumulation is the accumulation and reporting of the scheduling cost overhead of the task request for a service scheduled in a scheduling cycle by the network access point;

[0028] Determine the target area to which the task request is scheduled according to the scheduling strategy;

[0029] The task request is dispatched to a network access point corresponding to the target area.

[0030] Wherein, after receiving the task request in response and before determining the service corresponding to the task request, the method further comprises:

[0031] Determine whether the task request is forwarded by a network access point, and if so, send it to a corresponding edge server for processing; otherwise, perform a service operation corresponding to the task request.

[0032] The scheduling probability is determined by the central controller before the start of each scheduling cycle based on the scheduling cost accumulation of each service and the computing power network diagram, including:

[0033] Calculate the excess accumulation of scheduling cost for each service at the beginning of the next scheduling cycle according to the scheduling cost overhead accumulation corresponding to each service reported by each network access node and the preset long-term average scheduling cost budget;

[0034] The scheduling strategy for the next scheduling cycle is determined according to the excess accumulation of scheduling cost and the computing power network diagram.

[0035] Wherein, after the current scheduling period starts, the method further includes:

[0036] Monitor the task requests for each service schedule in the current scheduling cycle;

[0037] When it is monitored that a task request is scheduled for any service, determining a scheduling cost overhead of scheduling the task request;

[0038] When the current scheduling cycle ends, the scheduling cost overhead corresponding to each service is accumulated, the scheduling cost overhead accumulation corresponding to each service is obtained, and reported to the central controller.

[0039] In one embodiment, a task scheduling device for collaborative edge computing is provided, which is applied to a central controller in a multi-region edge interconnection network, and the device includes:

[0040] A receiving unit is configured to receive the scheduling cost accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service reported by each network access point in the multi-region edge interconnection network; the scheduling cost accumulation is the accumulation of scheduling cost of a task request scheduled by a network access point for a service in a scheduling cycle;

[0041] The processing unit is configured to perform, at the end of the current scheduling cycle, for each service of each network access point, respectively, according to the corresponding task arrival prediction information, resource allocation information and network status information to obtain a computing power network graph; and determine a scheduling strategy for the next scheduling cycle according to the scheduling cost overhead accumulation and the computing power network graph; the scheduling strategy is a scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network;

[0042] The sending unit is configured to send the scheduling policy determined for each service of each network access point to the corresponding network access point before the next scheduling period starts.

[0043] In one embodiment, a task scheduling device for collaborative edge computing is provided, which is applied to a network access point in any area of ​​a multi-area edge interconnection network, and the device includes:

[0044] A first determining unit is configured to determine, in response to receiving a task request, a service corresponding to the task request;

[0045] The acquisition unit is configured to execute and acquire the scheduling strategy corresponding to the service in the current scheduling period; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network; the scheduling probability is determined and issued by the central controller before the start of each scheduling period based on the scheduling cost overhead accumulation and the computing power network diagram of each service; the computing power network diagram is calculated based on the task arrival prediction information, resource allocation information and network status information reported by each network access point in the multi-area edge Internet; the scheduling cost overhead accumulation is the accumulation and reporting of the scheduling cost overhead of the network access point for scheduling task requests for a service in a scheduling period;

[0046] A second determining unit is configured to determine a target area to which the task request is scheduled according to the scheduling strategy;

[0047] The scheduling unit is configured to schedule the task request to a network access point corresponding to the target area.

[0048] In another embodiment, a task scheduling system for collaborative edge computing is provided, which is applied in a multi-region edge internet network, characterized in that the system includes:

[0049] A network access point, in response to receiving a task request, determines a service corresponding to the task request; obtains a scheduling policy corresponding to the service in a current scheduling period; determines a target area to which the task request is scheduled according to the scheduling policy; and sends the task request to a network access point corresponding to the target area;

[0050] A central controller receives the scheduling cost accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service reported by each network access point in the multi-region edge interconnection network; the scheduling cost accumulation is the accumulation of scheduling cost of scheduling task requests for a service by the network access point in a scheduling cycle; at the end of the current scheduling cycle, for each service of each network access point, respectively, a computing power network graph is calculated according to the corresponding task arrival prediction information, resource allocation information and network status information; and a scheduling strategy for the next scheduling cycle is determined according to the scheduling cost accumulation and the computing power network graph; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each region in the multi-region edge interconnection network; before the start of the next scheduling cycle, the scheduling strategy determined for each service of each network access point is issued to the corresponding network access point.

[0051] In another embodiment, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the task scheduling method for collaborative edge computing as described above is implemented.

[0052] In another embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the program is executed by a processor, the task scheduling method for collaborative edge computing as described above is implemented.

[0053] It can be seen from the above technical solution that in the above embodiment, at the end of each scheduling cycle, the scheduling strategy of each task is determined according to the scheduling cost overhead accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service obtained by each network access point, to guide the scheduling of task requests in the next scheduling cycle; this solution decomposes a problem measured from a long-term perspective into each task scheduling cycle, considers the scheduling cost in each scheduling cycle, and can achieve efficient collaborative edge computing under the premise of controlling the scheduling task cost to meet the long-term average scheduling cost budget. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0055] Figure 1 A schematic diagram of a task scheduling process for implementing collaborative edge computing in a network access point in an embodiment of the present application;

[0056] Figure 2 This is a schematic diagram of a task scheduling process for implementing collaborative edge computing on another network access point in an embodiment of the present application;

[0057] Figure 3 A schematic diagram of a flow chart for determining the scheduling cost accumulation corresponding to a service in an embodiment of the present application;

[0058] Figure 4 This is a schematic diagram of the task scheduling process of the central controller implementing collaborative edge computing in an embodiment of the present application;

[0059] Figure 5 A schematic diagram of a computing power network diagram calculated in an embodiment of the present application;

[0060] Figure 6 A schematic diagram of a process for determining a scheduling strategy for the next scheduling cycle for any task in an embodiment of the present application;

[0061] Figure 7 A schematic diagram of a weighted directed acyclic graph without intermediate nodes in an embodiment of the present application;

[0062] Figure 8 A schematic diagram of a weighted directed acyclic graph with intermediate nodes in an embodiment of the present application;

[0063] Fig. 9 This is a schematic diagram of a task scheduling system for collaborative edge computing in an embodiment of the present application;

[0064] Fig.10 This is a schematic diagram of the structure of a task scheduling device for collaborative edge computing in an embodiment of the present application;

[0065] Fig.11 This is a schematic diagram of the structure of another task scheduling device for collaborative edge computing in an embodiment of the present application;

[0066] Fig.12 A schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0067] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0068] The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe the order or precedence of the objects. It should be understood that the data used in this way can be interchangeable where appropriate, so that the implementation of the present invention described herein can be implemented in an order other than those illustrated or described herein, for example. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0069] The technical solution of the present invention is described in detail with specific embodiments below. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described in detail in some embodiments.

[0070] The edge collaboration implementation scheme in related technologies does not take scheduling costs into consideration. Task scheduling between edge nodes often generates additional cost overhead, mainly network cost. Obviously, cost is a key factor to consider, that is, it is necessary to enhance the user's service experience quality as much as possible under cost control. Given the changes in user task requests that are difficult to accurately predict and the long-term cost expectations that are set, how to achieve the best task scheduling in each scheduling cycle is the main challenge of collaborative edge computing.

[0071] In view of the above problems, an embodiment of the present application discloses a task scheduling method for collaborative edge computing, which is applied to a multi-region edge Internet network; at the end of each scheduling cycle, the scheduling strategy of each task is determined according to the accumulated scheduling cost overhead, task arrival prediction information, resource allocation information and network status information corresponding to each service obtained by each network access point, to guide the scheduling of task requests in the next scheduling cycle; this scheme decomposes a problem measured from a long-term perspective into each task scheduling cycle, considers the scheduling cost in each scheduling cycle, and can efficiently realize edge collaborative computing under the premise of controlling the scheduling task cost, and can solve the problem that when a multi-region edge Internet network faces user task requests with uneven temporal and spatial distribution, it is difficult to capture changes in the system state, resulting in the decision-making being unable to achieve a balance between short-term performance and long-term constraints.

[0072] The multi-region edge Internet includes multiple regions, such as multiple factories, multiple campuses, etc.; each region is deployed with a computing resource pool, and the networks of each region are connected by a wide area network. A total of multiple services are deployed in these regions, wherein one or more services are deployed in each computing resource pool, and certain resources are allocated to each deployed service; wherein service deployment and resource allocation are based on historical statistics or other strategies, which are not limited in the embodiments of the present application, but rather provide a method for optimally scheduling task requests arriving in each region in each scheduling cycle after completing service deployment and resource allocation.

[0073] The scheduling period in the embodiment of the present application can be configured by analyzing the historical request arrival of the corresponding service, or it can be adaptively adjusted through online learning. The scheduling period of each service can be the same or different because the scheduling of different services is independent of each other. In the embodiment of the present application, the same scheduling period is taken as an example below.

[0074] Users in each area initiate task requests to a service deployed in the edge Internet network at any time. The task request first arrives at the network access point of the corresponding area. It is necessary to determine the service corresponding to the task request. If the corresponding service is a service deployed in the edge network, the technical solution provided in the embodiment of the present application is used to process the task request. If the corresponding service is not a service deployed in the edge network, the processing solution in the related technology is used to process the task request.

[0075] See also Figure 1 , Figure 1 This is a schematic diagram of a task scheduling process for implementing collaborative edge computing in a network access point in an embodiment of the present application. The specific steps are:

[0076] Step 101, in response to receiving a task request, determining a service corresponding to the task request.

[0077] Step 102, obtaining the scheduling strategy corresponding to the service in the current scheduling period; wherein the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network.

[0078] Among them, the scheduling probability is determined and issued by the central controller before the start of each scheduling cycle based on the scheduling cost accumulation of each service and the computing network diagram; the computing network diagram is calculated based on the task arrival prediction information, resource allocation information and network status information reported by each network access point in the multi-region edge Internet; the scheduling cost accumulation is the accumulation and reporting of the scheduling cost of the scheduling task request of a service in a scheduling cycle by the network access point;

[0079] The task prediction information includes: the predicted number of task requests for the i-th service that arrive at the j-th network access point in the current scheduling period; wherein j is greater than 0 and not greater than the number of network access points in the multi-region edge Internet; i is greater than 0 and not greater than the number of services deployed in the multi-region edge Internet;

[0080] Take the example of network access point 1 obtaining task prediction information corresponding to service 1, that is, network access point 1 predicts the number of task requests corresponding to service 1 arriving in the current scheduling cycle;

[0081] The resource allocation information includes: the amount of computing resources allocated by the j-th edge computing resource pool to the i-th service; here, taking the deployment of one edge computing resource pool in one area as an example, the number of computing resource pools is the same as the number of network access points. If multiple edge computing resource pools are deployed in one area, the amount of allocated computing resources can be expressed based on the relationship between the specific edge computing resource pools and the service.

[0082] Take the example of network access point 1 obtaining the computing resources corresponding to the service; that is, the network access point determines the computing resources allocated to the service in the edge computing resource pool in the area where it is located;

[0083] The network status information includes: a network bandwidth between an mth region and an nth region, wherein m and n are different and both are greater than 0 and less than the number of regions in the multi-region edge Internet.

[0084] The network access point measures the network bandwidth between other network access points to indicate the network bandwidth between the area where the network access point is located and the areas where other network access points are located. The network bandwidth from the network access point to itself can be determined as the maximum bandwidth allowed.

[0085] Step 103: Determine the target area to which the task request is scheduled according to the scheduling policy.

[0086] Step 104: dispatch the task request to the network access point corresponding to the target area.

[0087] In this embodiment, at the end of each scheduling cycle, the network access point obtains the accumulated scheduling cost overhead, task arrival prediction information, resource allocation information and network status information corresponding to each service and reports them to the central processor, so that the central controller determines the scheduling strategy for each task before the start of the next scheduling cycle; when the network access point receives a task request, it schedules the task request according to the scheduling strategy of the service corresponding to the task request. This solution decomposes a problem measured from a long-term perspective into each task scheduling cycle, considers the scheduling cost in each scheduling cycle, and can efficiently realize edge collaborative computing under the premise of controlling the scheduling task cost, and can solve the problem that when multi-region edge interconnection networks face user task requests with uneven temporal and spatial distribution, it is difficult to capture changes in system status, resulting in the decision-making being unable to achieve a balance between short-term performance and long-term constraints.

[0088] In another embodiment, the central controller determines the scheduling probability according to the scheduling cost accumulation of each service and the computing power network graph before each scheduling period starts, including:

[0089] Calculate the excess accumulation of scheduling cost for each service corresponding to each network access point at the beginning of the next scheduling cycle according to the scheduling cost overhead accumulation corresponding to each service reported by each network access node and the preset long-term average scheduling cost budget;

[0090] The scheduling strategy for the next scheduling cycle is determined based on the excess accumulation of scheduling costs and the computing power network diagram.

[0091] Among them, the preset long-term average scheduling overhead cost refers to the average value of the task scheduling overhead over a longer period of time from a long-term perspective. The scheduling cycle is generally much smaller than the period here. When calculating the excess accumulation of scheduling cost, the proportional coefficient is used to scale it and the actual scheduling cost accumulation to the same time length.

[0092] See also Figure 2 , Figure 2 This is a schematic diagram of a task scheduling process for implementing collaborative edge computing on another network access point in an embodiment of the present application. The specific steps are:

[0093] Step 201: Receive a task request.

[0094] Step 202 , determining whether the task request is forwarded by a network access point, if yes, executing step 207 ; otherwise, executing step 203 .

[0095] Step 203: determine the service corresponding to the task request.

[0096] Step 204, obtaining the scheduling strategy corresponding to the service in the current scheduling period; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network.

[0097] The scheduling probability is determined and issued by the central controller before the start of each scheduling cycle based on the scheduling cost accumulation of each service and the computing power network diagram; the computing power network diagram is calculated based on the task arrival prediction information, resource allocation information and network status information reported by each network access point in the multi-region edge Internet; the scheduling cost accumulation is the accumulation and reporting of the scheduling cost of the task request for a service scheduled in a scheduling cycle by the network access point.

[0098] The task prediction information includes: the predicted number of task requests for the i-th service that arrive at the j-th network access point in the current scheduling period; wherein j is greater than 0 and not greater than the number of network access points in the multi-region edge Internet; i is greater than 0 and not greater than the number of services deployed in the multi-region edge Internet;

[0099] The resource allocation information includes: the amount of computing resources allocated by the j-th edge computing resource pool to the i-th service; here, taking the deployment of one edge computing resource pool in one area as an example, the number of computing resource pools is the same as the number of network access points. If multiple edge computing resource pools are deployed in one area, the amount of allocated computing resources can be expressed based on the relationship between the specific edge computing resource pools and the service.

[0100] The network status information includes: a network bandwidth between an mth region and an nth region, wherein m and n are different and both are greater than 0 and less than the number of regions in the multi-region edge Internet.

[0101] Step 205: Determine the target area to which the task request is scheduled according to the scheduling policy.

[0102] Step 206: dispatch the task request to the network access point corresponding to the target area. This process ends.

[0103] Step 207: Send to the corresponding edge server for processing.

[0104] In this embodiment, at the end of each scheduling cycle, the network access point obtains the accumulated scheduling cost overhead, task arrival prediction information, resource allocation information and network status information corresponding to each service and reports them to the central processor, so that the central controller determines the scheduling strategy for each task before the start of the next scheduling cycle; when the network access point receives a task request, the task request forwarded by the network access point is directly sent to the edge server for processing; for the task request sent by the user, the task request is scheduled according to the scheduling strategy of the service corresponding to the task request. This solution decomposes a problem measured from a long-term perspective into each task scheduling cycle, considers the scheduling cost in each scheduling cycle, and can achieve efficient collaborative edge computing under the premise of controlling the scheduling task cost, and can solve the problem that when a multi-region edge interconnected network faces user task requests with uneven temporal and spatial distribution, it is difficult to capture changes in the system state, resulting in the decision-making being unable to achieve a balance between short-term performance and long-term constraints.

[0105] See also Figure 3 , Figure 3 This is a flow chart of determining the scheduling cost accumulation corresponding to a service in an embodiment of the present application. The specific steps are:

[0106] Step 301: After the current scheduling cycle starts, monitor the task requests for each service scheduling in the current scheduling cycle.

[0107] Step 302: When it is monitored that a task request is scheduled for any service, a scheduling cost overhead of scheduling the task request is determined.

[0108] Step 303: When the current scheduling cycle ends, the scheduling cost overhead corresponding to each service is accumulated to obtain the scheduling cost overhead accumulation corresponding to each service, and report it to the central controller.

[0109] In this embodiment, the network access point accumulates the scheduling cost overhead of the scheduling task request for a task in a scheduling cycle, and reports it to the central controller.

[0110] See also Figure 4 , Figure 4 This is a schematic diagram of the task scheduling process of the central controller to implement collaborative edge computing in an embodiment of the present application. The specific steps are:

[0111] Step 401, receiving the scheduling cost accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service reported by each network access point in the multi-region edge interconnection network.

[0112] The scheduling cost overhead accumulation is the accumulation of scheduling cost overhead of scheduling task requests of a service in a scheduling cycle by a network access point;

[0113] The process of network access point obtaining scheduling cost accumulation is as follows:

[0114] After the current scheduling cycle starts, monitor the task requests for each service scheduling in the current scheduling cycle;

[0115] When it is detected that a task request is scheduled for any service, a scheduling cost overhead of scheduling the task request is determined;

[0116] When the current scheduling cycle ends, the scheduling cost overhead corresponding to each service is accumulated to obtain the scheduling cost overhead accumulation corresponding to each service.

[0117] Step 402, at the end of the current scheduling cycle, a computing power network diagram is calculated for each service of each network access point according to the corresponding task arrival prediction information, resource allocation information and network status information; and a scheduling strategy for the next scheduling cycle is determined according to the scheduling cost overhead accumulation and the computing power network diagram; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network.

[0118] See also Figure 5 , Figure 5 A schematic diagram of a computing power network diagram calculated in an embodiment of the present application. Figure 5 Take the computing power network diagram corresponding to service S as an example; It indicates the number of requests arriving in the next scheduling period predicted by the nth network access point based on the historical request arrival situation of the service. Indicates the amount of computing resources allocated by the edge resource pool in the nth region to service S; Represents the network bandwidth between the mth region and the nth region, where m and n are different and both are greater than 0 and not greater than 4, taking 4 as the number of regions in the multi-region edge Internet (which is also equivalent to the number of network access points) as an example.

[0119] In step 402, the scheduling strategy for the next scheduling cycle is determined according to the scheduling cost accumulation and the computing power network diagram, including:

[0120] Calculate the excess accumulation of scheduling cost for each service at the beginning of the next scheduling cycle according to the scheduling cost overhead accumulation corresponding to each service reported by each network access node and the preset long-term average scheduling cost budget;

[0121] For example: Assume that there are M network access points in total. For service S, in the t-th scheduling cycle, the m-th wireless access point dispatches each request from the user for service S to each network access point according to the scheduling strategy issued, and then forwards it to the corresponding edge computing server. After forwarding each task request, the network access point will monitor the cost overhead generated by forwarding the task request. When the t-th policy cycle ends, each network access point will accumulate the scheduling cost overhead generated by the task request for scheduling service S in this cycle, obtain the scheduling cost overhead accumulation and report it to the central controller.

[0122] Assume that the excess scheduling cost for service S at the beginning of the tth period is accumulated to Q S (t), the scheduling cost of the mth network access point in this period is accumulated as The length of a scheduling cycle is T S , the long-term average scheduling overhead cost budget of the service is Then the calculation method for the excess accumulation of scheduling cost of the service at the beginning of the next scheduling cycle is:

[0123]

[0124] The scheduling strategy for the next scheduling cycle is determined based on the excess accumulation of scheduling costs and the computing power network diagram.

[0125] The process of determining the scheduling strategy for the next scheduling cycle is implemented based on a modified genetic algorithm. Taking service S as an example, assume that the excess scheduling overhead accumulated at the end of the previous scheduling cycle is Q S (t), then the computing network graph of service S and Q S (t) is used as the input of the improved genetic algorithm to output the scheduling strategy of service S in the next scheduling cycle. Taking 4 regions as an example, the scheduling strategy of service S is in the form of:

[0126]

[0127] in, represents the probability that a network access point in region m will schedule each request from a user for service S to region n. Therefore, the sum of the probabilities in each row is equal to 1.

[0128] Step 403: before the next scheduling cycle starts, the scheduling strategy determined for each service of each network access point is sent to the corresponding network access point.

[0129] In this embodiment, the central controller determines the scheduling strategy for each service by reporting the scheduling cost overhead accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service through the network access point at the end of each scheduling cycle; and sends it to the network access point before the start of the next scheduling cycle for scheduling task requests. This solution decomposes a problem measured from a long-term perspective into each task scheduling cycle, considers the scheduling cost in each scheduling cycle, and can achieve efficient collaborative edge computing under the premise of controlling the scheduling task cost, and can solve the problem that when multi-region edge Internet networks face user task requests with uneven temporal and spatial distribution, it is difficult to capture changes in system status, resulting in the decision-making being unable to achieve a balance between short-term performance and long-term constraints.

[0130] See also Figure 6 , Figure 6 This is a flow chart of determining the scheduling strategy for the next scheduling cycle for any task in an embodiment of the present application. The specific steps are:

[0131] Step 601, randomly generate an initial region category allocation solution; wherein the region categories include: source node, sink node and isolated node; wherein the source node has only out-degree but no in-degree; the sink node has only in-degree but no out-degree; and the isolated node has neither out-degree nor in-degree.

[0132] Initialize the iteration counter to 0 and the maximum number of iterations to N iter , crossover probability P cross and mutation probability P mutation .

[0133] Corresponding to each area, the area belonging to the source node means that the area has a strategy to dispatch task requests to other areas, and does not receive task requests dispatched from other areas; the area belonging to the sink node means that the area has a strategy to receive task requests dispatched from other areas, but does not dispatch task requests to other areas; the area belonging to the isolated node means that the area does not receive requests from other areas and dispatches requests to other areas, that is, all out-degree and all in-degree weights (i.e., probabilities) are 0. Once there are other types of nodes, the corresponding scheduling decision must be suboptimal.

[0134] Assuming there are M regions in total, for service S, the generated initial region category allocation solution can be: in Where 0 represents a source node, 1 represents a sink node, and 2 represents an isolated node. but And if O s If there is an element equal to 1, there must be an element equal to 0. That is, as long as there is a sink node, there must be a source node.

[0135] The solution sought must satisfy the following three conditions, that is, the optimal solution is obtained under these three conditions. The role of these three conditions is to reduce the search space for the optimal solution:

[0136] First, the optimal traffic scheduling strategy graph is a weighted directed acyclic graph with no intermediate nodes;

[0137] See also Figure 7 , Figure 7 This is a schematic diagram of a weighted directed acyclic graph without any intermediate nodes in an embodiment of the present application. Figure 7 The four regions are four nodes, and there is no intermediate node; the sink node refers to the edge with only in-degree, that is, all the edges connected to it are pointing to it, such as Figure 7 The node corresponding to region 3 in the middle; the source node is the only out-degree edge, that is, all the edges connected to it are from it to other nodes, such as Figure 7 The nodes corresponding to regions 1 and 2 in the figure. The points corresponding to region 4 are isolated nodes.

[0138] Weighting refers to the weights marked on the lines in the figure, which is the scheduling probability. For example, the weight from area 1 to area 3 is P 13 , the weight of area 2 to area 3 is P 23 ; In a directed acyclic graph, directed means that the lines between nodes are directed edges, and an acyclic graph means that starting from any vertex, you cannot return to the starting vertex by moving along the directed edges, that is, it will not form a loop. An intermediate node means that the node has a predecessor node (that is, other vertices are connected to the vertex through edges pointing to the vertex) and a successor node (that is, the vertex is connected to other vertices through edges pointing to other vertices.)

[0139] See also Figure 8 , Figure 8 This is a schematic diagram of a weighted directed acyclic graph with intermediate nodes in an embodiment of the present application. Figure 8 The four regions in the middle are used as four nodes, and the node corresponding to region 3 is the middle node.

[0140] The traffic scheduling strategy graph here can be converted into a scheduling strategy, where only the probability corresponding to the weighted edge is greater than 0, and the probability of staying locally is 1 minus the sum of all outgoing edge weights, and the rest are equal to 0.

[0141] Second, for the task scheduling strategy of a certain service, the area where the service is not deployed can only be the source node; if but

[0142] Third, for a region belonging to a source node, if the service is deployed in the region, the sum of the weights of all outgoing edges cannot exceed 1, otherwise, the sum of the weights of all outgoing edges is equal to 1; this can be expressed as and but if and but

[0143] Step 602, for each area category allocation solution, solve the optimal scheduling strategy and the corresponding objective function value under the current area category allocation solution; the objective function value is determined by the Lyapunov shift plus penalty function, and the Lyapunov shift plus penalty function incorporates the excess accumulation of scheduling cost into the delay penalty.

[0144] Taking service S as an example, the objective function defined for service S is:

[0145]

[0146] Constraints

[0147] Among them, P s represents the scheduling strategy (scheduling probability) to be determined for service S; V is the penalty coefficient, which is a hyperparameter. When the V value is larger, it means that the delay requirement is stricter, and vice versa, the task scheduling cost constraint is stricter; T s (P s ) indicates that under the condition of a given computing power network graph, the scheduling probability P is adopted S Since the distribution of task request quantity, resource distribution and task scheduling probability are known, the expected number of tasks scheduled to the edge servers corresponding to each network access point can be derived, thereby calculating the expected task processing delay. s (P s ) indicates that under the condition of a given computing power network graph, the scheduling strategy (scheduling probability) P is adopted S The expected task scheduling cost when represents the long-term average scheduling overhead cost budget set for service S; the constraint condition means that the sum of each row in the policy matrix is ​​equal to 1.

[0148] Furthermore, the harmony search algorithm is used to solve the optimal task scheduling strategy and the corresponding objective function value that can be achieved under a given area category allocation scheme. The input is a regional category allocation solution. Taking 4 areas as an example, the regional category allocation solution is a list of length 4, and each value in the list is 0, 1, or 2, representing 3 different categories. The output is the optimal task scheduling strategy under the regional category allocation solution. and the corresponding objective function value.

[0149] Step 603, according to each area category allocation solution and the corresponding optimal scheduling strategy and objective function value, the fitness corresponding to the area category allocation solution is obtained, and the area category allocation solutions corresponding to the fitness values ​​greater than the preset fitness values ​​are selected to form an optimized area category allocation solution group.

[0150] Step 604 , determine whether the maximum number of iterations or the preset optimization condition is reached, if so, execute step 607 ; otherwise, execute step 605 .

[0151] Step 605: Based on the optimized area category allocation solution, a first extended area category allocation group is generated using a crossover probability and a crossover method.

[0152] According to the given crossover probability and crossover method, a new chromosome set is generated according to the current chromosome set to increase the diversity of the regional category allocation scheme. The crossover location is only for the area where the service is deployed, because the area where the service is not deployed can only be the source node, and it needs to dispatch all external user requests for the service to other areas belonging to the sink node for processing, that is, it cannot process the service requests locally.

[0153] Step 606, based on the first extended area category allocation solution group, using mutation probability and mutation method to generate a second extended area category allocation solution group; execute step 602;

[0154] Traverse the first extended area category assignment solution group, based on the initialization of the mutation probability P mutation Generate a random number Random between 0 and 1. If Random>P mutation , then the gene (allocation solution) will not be mutated; otherwise, a position (between 1 and M) is randomly selected, and the element at the corresponding position of this solution is modified to any value in {0, 1, 2} and different from the original value of the element; then, it is determined whether the extended allocation solution meets the above three conditions. If so, it is retained; otherwise, the mutated solution is discarded and regenerated. The area involved in the mutation operation is only the area where the service is deployed.

[0155] Step 607, output the optimal scheduling strategy as the scheduling strategy determined according to the scheduling cost overhead accumulation and the computing power network diagram; wherein the optimal scheduling strategy is the strategy corresponding to the regional category allocation solution with the maximum fitness in the optimized regional category allocation solution group.

[0156] For a given regional category solution, a corresponding traffic offloading strategy graph can be constructed, but it is not weighted. The following method is to find the optimal weight value of each edge (corresponding to the task scheduling probability).

[0157] In step 602, when solving the optimal scheduling strategy under the current area category allocation, it is obtained through the harmony search algorithm. The specific solution process is as follows:

[0158] The first step is to construct optimization variables according to the categories of each region, and determine the upper and lower bounds of each optimization variable; initialize the number of iterations, the size of the harmonic memory bank HMS, the harmonic memory retention probability HMCR, the pitch adjustment probability PAR, and the adjustment bandwidth BW;

[0159] The optimization variable is the probability value corresponding to the weight of the edge of the constructed unweighted traffic offloading strategy graph. The upper bound of each optimization variable is 1 and the lower bound is 0.

[0160] The second step is to initialize the HMS harmonies according to the value range of each optimization variable; that is, to initialize a set of initial solutions consisting of the values ​​of the optimization variables;

[0161] Step 3: At the beginning of a new iteration, traverse each optimization variable and generate a random number between 0 and 1 for each optimization variable. If the random number is less than HMCR, randomly select the value of the variable in the harmony memory library, otherwise randomly select a value within the allowed range of the variable.

[0162] Step 4: perturb the variable with a bandwidth of BW using probability PAR, and regulate the perturbed variable to be within the allowed value range;

[0163] Step 5: Convert the new harmony into a task scheduling strategy solution, and solve the value of the expected objective function that can be achieved by the harmony to determine whether the new solution is better than the worst solution in the harmony memory library; if it is better, replace the worst solution in the harmony memory library with the new solution.

[0164] Assume that the transformed task scheduling strategy solution is P S , then the corresponding objective function is

[0165] Step 6: Determine whether the stopping criteria are met. If yes, execute step 7; otherwise, execute step 3;

[0166] The stopping criterion is whether the current number of iterations reaches the preset number of iterations.

[0167] Step 7: Return the task scheduling strategy corresponding to the optimal harmony in the harmony memory library and the objective function value corresponding to the scheduling strategy.

[0168] So far, the entire process of the scheduling strategy is fully described.

[0169] See also Fig. 9 , Fig. 9This is a schematic diagram of a task scheduling system for collaborative edge computing in an embodiment of the present application. The system includes: a central controller and a network access point;

[0170] The network access point, in response to receiving the task request, determines the service corresponding to the task request; obtains the scheduling policy corresponding to the service in the current scheduling period; determines the target area to which the task request is scheduled according to the determined scheduling policy; and sends the task request to the network access point corresponding to the target area;

[0171] A central controller receives the scheduling cost accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service reported by each network access point in the multi-region edge interconnection network; wherein the scheduling cost accumulation is the accumulation of scheduling cost of scheduling task requests of a service by the network access point in a scheduling cycle; at the end of the current scheduling cycle, a computing power network diagram is calculated for each service of each network access point according to the corresponding task arrival prediction information, resource allocation information and network status information; and a scheduling strategy for the next scheduling cycle is determined according to the scheduling cost accumulation and the computing power network diagram; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each region in the multi-region edge interconnection network; before the start of the next scheduling cycle, the scheduling strategy determined for each service of each network access point is sent to the corresponding network access point.

[0172] In another example,

[0173] The central controller determines the specific implementation of the scheduling strategy for the next scheduling cycle based on the scheduling cost accumulation and computing power network diagram:

[0174] Calculate the excess accumulation of scheduling cost for each service at the beginning of the next scheduling cycle according to the scheduling cost overhead accumulation corresponding to each service reported by each network access node and the preset long-term average scheduling cost budget;

[0175] The scheduling strategy for the next scheduling cycle is determined based on the excess accumulation of scheduling costs and the computing power network diagram.

[0176] In another example, for any service, the central controller determines the scheduling strategy for the next scheduling cycle according to the excess accumulation of scheduling cost and the computing power network diagram, including the steps of:

[0177] A. Randomly generate the initial region category allocation solution group; the region categories include: source node, sink node and isolated node; the source node has only out-degree but no in-degree; the sink node has only in-degree but no out-degree; the isolated node has neither out-degree nor in-degree;

[0178] B. For each area category allocation solution, solve the optimal scheduling strategy and the corresponding objective function value under the current area category allocation solution; the objective function value is determined by the Lyapunov shift plus penalty function, and the Lyapunov shift plus penalty function incorporates the excess accumulation of scheduling costs into the delay penalty;

[0179] C. According to each regional category allocation solution and the corresponding optimal scheduling strategy and objective function value, the fitness corresponding to the regional category allocation solution is obtained, and the regional category allocation solutions corresponding to the fitness values ​​greater than the preset fitness values ​​are selected to form an optimized regional category allocation solution group;

[0180] D. Determine whether the maximum number of iterations or the preset optimization condition is reached. If yes, execute step G; otherwise, execute step E;

[0181] E. Based on the optimized regional category allocation solution group, a first extended regional category allocation group is generated using a crossover probability and a crossover method;

[0182] F. Based on the first extended area category allocation solution group, generate a second extended area category allocation solution group using mutation probability and mutation method; execute step B;

[0183] G. Output the optimal scheduling strategy as the scheduling strategy determined according to the scheduling cost accumulation and the computing power network diagram; the optimal scheduling strategy is the strategy corresponding to the regional category allocation solution with the highest fitness in the optimized regional category allocation solution group.

[0184] In another example,

[0185] The central controller,solves the optimal scheduling strategy under the current area category allocation,through the harmony search algorithm.

[0186] In another example,

[0187] After receiving the task request, the network access point determines whether the task request is forwarded by the network access point before determining the service corresponding to the task request. If so, it sends it to the corresponding edge server for processing; otherwise, it executes the service operation corresponding to the task request.

[0188] In another example,

[0189] The network access point, after the current scheduling cycle starts, monitors the task requests for each service scheduling in the current scheduling cycle; when it is monitored that the task request is scheduled for any service, the scheduling cost overhead of the scheduling task request is determined; when the current scheduling cycle ends, the scheduling cost overhead corresponding to each service is accumulated, the scheduling cost overhead corresponding to each service is obtained, and the accumulated scheduling cost overhead is reported to the central controller.

[0190] In the embodiment of the present application, a task scheduling strategy for collaborative edge computing is developed under the guidance of Lyapunov optimization theory to address the spatiotemporal non-uniformity and time-varying distribution of user requests in multiple regions. This embodiment applies the solution of a random optimization problem measured from a long-term perspective to each scheduling cycle without requiring future system status information or even the probability distribution of random events.

[0191] A computing power network graph model was constructed for service deployment, resource allocation, network status, and task arrival prediction in multi-region edge Internet networks. Based on this model, an improved genetic algorithm and harmony search algorithm were used to solve the optimal task scheduling decision in each scheduling cycle.

[0192] It is suitable for systems in time-varying environments, realizes load balancing of multi-region edge Internet networks, reduces the overall response time of user tasks, and ensures that the long-term average cost of task scheduling is controlled.

[0193] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present disclosure, and will not be described in detail here.

[0194] Based on the same inventive concept, the present application also provides a task scheduling device for collaborative edge computing, which is applied to a central controller in a multi-region edge interconnection network. Fig.10 , Fig.10 This is a schematic diagram of the structure of a task scheduling device for collaborative edge computing in an embodiment of the present application. The device includes:

[0195] The receiving unit 1001 is configured to receive the scheduling cost accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service reported by each network access point in the multi-region edge interconnection network; the scheduling cost accumulation is the accumulation of scheduling cost of the network access point for scheduling task requests for a service in a scheduling cycle;

[0196] The processing unit 1002 is configured to execute, at the end of the current scheduling cycle, for each service of each network access point, respectively, according to the corresponding task arrival prediction information, resource allocation information and network status information, to obtain a computing power network graph; and determine a scheduling strategy for the next scheduling cycle according to the scheduling cost overhead accumulation and the computing power network graph; the scheduling strategy is a scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network;

[0197] The sending unit 1003 is configured to send the scheduling policy determined for each service of each network access point to the corresponding network access point before the next scheduling period starts.

[0198] In another example,

[0199] Processing unit 1002 is configured to execute, when determining the scheduling strategy for the next scheduling cycle based on the scheduling cost accumulation and the computing power network diagram, calculate the excess scheduling cost accumulation of each service at the beginning of the next scheduling cycle based on the scheduling cost accumulation corresponding to each service reported by each network access node and the preset long-term average scheduling cost budget; and determine the scheduling strategy for the next scheduling cycle based on the excess scheduling cost accumulation and the computing power network diagram.

[0200] In another example,

[0201] The processing unit 1002 is configured to execute a scheduling strategy for determining a next scheduling period according to the excess accumulation of scheduling cost and the computing power network, including:

[0202] A. Randomly generate the initial region category allocation solution group; the region categories include: source node, sink node and isolated node; the source node has only out-degree but no in-degree; the sink node has only in-degree but no out-degree; the isolated node has neither out-degree nor in-degree;

[0203] B. For each area category allocation solution, solve the optimal scheduling strategy and the corresponding objective function value under the current area category allocation solution; the objective function value is determined by the Lyapunov shift plus penalty function, and the Lyapunov shift plus penalty function incorporates the excess accumulation of scheduling costs into the delay penalty;

[0204] C. According to each regional category allocation solution and the corresponding optimal scheduling strategy and objective function value, the fitness corresponding to the regional category allocation solution is obtained, and the regional category allocation solutions corresponding to the fitness values ​​greater than the preset fitness values ​​are selected to form an optimized regional category allocation solution group;

[0205] D. Determine whether the maximum number of iterations or the preset optimization condition is reached. If yes, execute step G; otherwise, execute step E;

[0206] E. Based on the optimized regional category allocation solution group, a first extended regional category allocation group is generated using a crossover probability and a crossover method;

[0207] F. Based on the first extended area category allocation solution group, generate a second extended area category allocation solution group using mutation probability and mutation method; execute step B;

[0208] G. Output the optimal scheduling strategy as the scheduling strategy determined according to the scheduling cost accumulation and the computing power network diagram; the optimal scheduling strategy is the strategy corresponding to the regional category allocation solution with the highest fitness in the optimized regional category allocation solution group.

[0209] In another example,

[0210] The processing unit 1002 is configured to execute the optimal scheduling strategy under the current area category allocation by using the harmony search algorithm.

[0211] In the embodiment of the present application, another task scheduling device for collaborative edge computing is provided, which is applied to a network access point in a multi-region edge interconnection network. Fig.11 , Fig.11 This is a schematic diagram of another task scheduling device for collaborative edge computing in an embodiment of the present application. The device includes:

[0212] The first determining unit 1101 is configured to determine the service corresponding to the task request in response to receiving the task request;

[0213] The acquisition unit 1102 is configured to execute and acquire the scheduling strategy corresponding to the service in the current scheduling cycle; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network; the scheduling probability is determined and issued by the central controller before the start of each scheduling cycle based on the scheduling cost overhead accumulation and the computing power network diagram of each service; the computing power network diagram is calculated based on the task arrival prediction information, resource allocation information and network status information reported by each network access point in the multi-area edge Internet; the scheduling cost overhead accumulation is the accumulation and reporting of the scheduling cost overhead of the network access point for scheduling task requests for a service in a scheduling cycle;

[0214] The second determining unit 1103 is configured to determine the target area to which the task request is scheduled according to the scheduling strategy;

[0215] The scheduling unit 1104 is configured to schedule the task request to a network access point corresponding to the target area.

[0216] In another example,

[0217] The first determining unit 1101 is configured to, in response to receiving a task request, determine whether the task request is forwarded by a network access point, and if so, send it to a corresponding edge server for processing; otherwise, determine a service corresponding to the task request.

[0218] In another example, the scheduling probability is determined by the central controller before each scheduling cycle starts based on the scheduling cost accumulation of each service and the computing power network graph, including:

[0219] The excess accumulation of scheduling cost for each service corresponding to each network access point at the beginning of the next scheduling cycle is calculated based on the scheduling cost overhead accumulation corresponding to each service reported by each network access node and the preset long-term average scheduling cost budget;

[0220] The scheduling strategy for the next scheduling cycle is determined based on the excess accumulation of scheduling costs and the computing power network.

[0221] In another embodiment, the device further comprises:

[0222] The monitoring unit 1105 is configured to monitor the task request for each service scheduling in the current scheduling cycle after the current scheduling cycle starts; when it is monitored that the task request is scheduled for any service, determine the scheduling cost overhead of the scheduling task request;

[0223] The reporting unit 1106 is configured to accumulate the scheduling cost corresponding to each service when the current scheduling cycle ends, obtain the scheduling cost accumulation corresponding to each service, and report it to the central controller.

[0224] The units in the above-mentioned embodiments may be integrated into one body or deployed separately; they may be combined into one unit or further divided into multiple sub-units.

[0225] In another embodiment, an electronic device is also provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, a task scheduling method for collaborative edge computing can be implemented.

[0226] In another embodiment, a computer-readable storage medium is also provided, on which computer instructions are stored, and when the instructions are executed by a processor, a task scheduling method for collaborative edge computing can be implemented.

[0227] Fig.12 Schematic diagram of the physical structure of an electronic device provided by an embodiment of the present invention. Fig.12 As shown, the electronic device may include: a processor (Processor) 1210, a communication interface (Communications Interface) 1220, a memory (Memory) 1230 and a communication bus 1240, wherein the processor 1210, the communication interface 1220, and the memory 1230 communicate with each other through the communication bus 1240. The processor 1210 may call the logic instructions in the memory 1230 to execute the following method:

[0228] Applied to the central controller in multi-region edge interconnection networks, including:

[0229] Receive the scheduling cost accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service reported by each network access point in the multi-region edge interconnection network; the scheduling cost accumulation is the accumulation of scheduling cost of scheduling task requests of a service by the network access point in a scheduling cycle;

[0230] At the end of the current scheduling cycle, the computing power network diagram is calculated for each service of each network access point based on the corresponding task arrival prediction information, resource allocation information and network status information; and the scheduling strategy for the next scheduling cycle is determined based on the scheduling cost overhead accumulation and the computing power network; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network;

[0231] Before the next scheduling period starts, the scheduling policy determined for each service of each network access point is sent to the corresponding network access point.

[0232] Applied to a network access point in any area of ​​a multi-area edge interconnection network, the method includes:

[0233] In response to receiving the task request, determining a service corresponding to the task request;

[0234] Obtain the scheduling strategy corresponding to the service in the current scheduling cycle; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-region edge interconnection network; the scheduling probability is determined and issued by the central controller before the start of each scheduling cycle based on the scheduling cost overhead accumulation of each service and the computing power network diagram; the computing power network diagram is calculated based on the task arrival prediction information, resource allocation information and network status information reported by each network access point in the multi-region edge Internet; the scheduling cost overhead accumulation is the accumulation and reporting of the scheduling cost overhead of the network access point for scheduling task requests for a service in a scheduling cycle;

[0235] Determine the target area to which the task request is scheduled according to the scheduling strategy;

[0236] Dispatch the task request to the network access point corresponding to the target area.

[0237] In addition, the logic instructions in the above-mentioned memory 1230 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0238] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0239] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0240] The flow chart and block diagram in the accompanying drawings of the present application show the possible architecture, function and operation of the system, method and computer program product according to the various embodiments disclosed in the present application. In this regard, each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the above-mentioned module, program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the box can also occur in the order of the standards in different figures. For example, the boxes represented by two connections can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0241] Those skilled in the art will appreciate that the features described in the various embodiments and / or claims disclosed in this application may be combined and / or combined in a variety of ways, even if such combinations and / or combinations are not explicitly described in this application. In particular, without departing from the spirit and teachings of this application, the features described in the various embodiments and / or claims of this application may be combined and / or combined in a variety of ways, and all of these combinations and / or combinations fall within the scope disclosed in this application.

[0242] Specific embodiments are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core ideas, and is not used to limit the present application. For those skilled in the art, changes can be made in the specific implementation methods and application scopes according to the ideas, spirits and principles of the present invention, and any modifications, equivalent substitutions, improvements, etc. made therein should be included in the scope of protection of this application.

Claims

1. A task scheduling method for collaborative edge computing, applied to a central controller in a multi-region edge interconnection network, characterized in that: The method comprises: Receiving the scheduling cost accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service reported by each network access point in the multi-region edge interconnection network; the scheduling cost accumulation is the accumulation of scheduling cost of scheduling task requests of a service by the network access point in a scheduling cycle; At the end of the current scheduling cycle, for each service of each network access point, a computing power network graph is calculated according to the corresponding task arrival prediction information, resource allocation information and network status information; and a scheduling strategy for the next scheduling cycle is determined according to the scheduling cost overhead accumulation and the computing power network graph; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network; Before the next scheduling period starts, the scheduling strategy determined for each service of each network access point is sent to the corresponding network access point.

2. The method according to claim 1, characterized in that The determining of the scheduling strategy for the next scheduling cycle according to the scheduling cost accumulation and the computing power network graph includes: The excess scheduling cost of each service is accumulated at the beginning of the next scheduling cycle according to the scheduling cost overhead accumulation corresponding to each service reported by each network access node and the preset long-term average scheduling cost budget; The scheduling strategy for the next scheduling cycle is determined according to the excess accumulation of scheduling cost and the computing power network diagram.

3. The method according to claim 1, characterized in that For any service, determining the scheduling strategy for the next scheduling cycle according to the excess accumulation of scheduling cost and the computing power network diagram comprises the steps of: A. Randomly generate an initial region category allocation solution group; wherein the region categories include: source node, sink node and isolated node; wherein the source node has only out-degree but no in-degree; the sink node has only in-degree but no out-degree; the isolated node has neither out-degree nor in-degree; B. For each area category allocation solution, solving the optimal scheduling strategy and the corresponding objective function value under the current area category allocation solution; the objective function value is determined by a Lyapunov shift plus penalty function, and the Lyapunov shift plus penalty function incorporates excess accumulation of scheduling costs into delay penalties; C. According to each regional category allocation solution and the corresponding optimal scheduling strategy and objective function value, the fitness corresponding to the regional category allocation solution is obtained, and the regional category allocation solutions corresponding to the fitness values ​​greater than the preset fitness values ​​are selected to form an optimized regional category allocation solution group; D. Determine whether the maximum number of iterations or the preset optimization condition is reached. If yes, execute step G; otherwise, execute step E; E. Based on the optimized area category allocation solution group, a first extended area category allocation group is generated using a crossover probability and a crossover method; F. Based on the first extended area category allocation solution group, generate a second extended area category allocation solution group using mutation probability and mutation method; execute step B; G. Output the optimal scheduling strategy as the scheduling strategy for the next scheduling cycle determined according to the accumulated scheduling cost and the computing power network diagram; the optimal scheduling strategy is the strategy corresponding to the regional category allocation solution with the maximum fitness in the optimized regional category allocation solution group.

4. The method according to claim 3, characterized in that The optimal scheduling strategy under the current area category allocation is solved by using a harmony search algorithm.

5. A task scheduling method for collaborative edge computing, applied to a network access point in any area of ​​a multi-area edge interconnection network, characterized in that: The method comprises: In response to receiving the task request, determining a service corresponding to the task request; Obtaining a scheduling strategy corresponding to the service in the current scheduling cycle; the scheduling strategy is a scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network; the scheduling probability is determined and issued by the central controller before the start of each scheduling cycle based on the scheduling cost overhead accumulation and the computing power network diagram of each service; the computing power network diagram is calculated based on the task arrival prediction information, resource allocation information and network status information reported by each network access point in the multi-area edge Internet; the scheduling cost overhead accumulation is the accumulation and reporting of the scheduling cost overhead of the task request for a service scheduled in a scheduling cycle by the network access point; Determine the target area to which the task request is scheduled according to the scheduling strategy; The task request is dispatched to a network access point corresponding to the target area.

6. The method according to claim 5, characterized in that After receiving the task request in response and before determining the service corresponding to the task request, the method further comprises: Determine whether the task request is forwarded by a network access point, and if so, send it to a corresponding edge server for processing; otherwise, perform a service operation corresponding to the task request.

7. The method according to claim 5, characterized in that The scheduling probability is determined by the central controller before the start of each scheduling cycle based on the scheduling cost accumulation of each service and the computing power network diagram, including: Calculate the excess accumulation of scheduling cost for each service at the beginning of the next scheduling cycle according to the scheduling cost overhead accumulation corresponding to each service reported by each network access node and the preset long-term average scheduling cost budget; The scheduling strategy for the next scheduling cycle is determined according to the excess accumulation of scheduling cost and the computing power network diagram.

8. The method according to claim 5, characterized in that After the current scheduling period starts, the method further includes: Monitor the task requests for each service schedule in the current scheduling cycle; When it is monitored that a task request is scheduled for any service, determining a scheduling cost overhead of scheduling the task request; When the current scheduling cycle ends, the scheduling cost overhead corresponding to each service is accumulated, the scheduling cost overhead accumulation corresponding to each service is obtained, and reported to the central controller.

9. A task scheduling device for collaborative edge computing, applied to a central controller in a multi-region edge internet network, characterized in that: The device comprises: A receiving unit is configured to receive the scheduling cost accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service reported by each network access point in the multi-region edge interconnection network; the scheduling cost accumulation is the accumulation of scheduling cost of a task request scheduled by a network access point for a service in a scheduling cycle; The processing unit is configured to perform, at the end of the current scheduling cycle, for each service of each network access point, respectively, according to the corresponding task arrival prediction information, resource allocation information and network status information to obtain a computing power network graph; and determine a scheduling strategy for the next scheduling cycle according to the scheduling cost overhead accumulation and the computing power network graph; the scheduling strategy is a scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network; The sending unit is configured to send the scheduling policy determined for each service of each network access point to the corresponding network access point before the next scheduling period starts.

10. A task scheduling device for collaborative edge computing, applied to a network access point in any area of ​​a multi-area edge internet network, characterized in that: The device comprises: A first determining unit is configured to determine, in response to receiving a task request, a service corresponding to the task request; The acquisition unit is configured to execute and acquire the scheduling strategy corresponding to the service in the current scheduling period; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each area in the multi-area edge interconnection network; the scheduling probability is determined and issued by the central controller before the start of each scheduling period based on the scheduling cost overhead accumulation and the computing power network diagram of each service; the computing power network diagram is calculated based on the task arrival prediction information, resource allocation information and network status information reported by each network access point in the multi-area edge Internet; the scheduling cost overhead accumulation is the accumulation and reporting of the scheduling cost overhead of the network access point for scheduling task requests for a service in a scheduling period; A second determining unit is configured to determine a target area to which the task request is scheduled according to the scheduling strategy; The scheduling unit is configured to schedule the task request to a network access point corresponding to the target area.

11. A task scheduling system for collaborative edge computing, applied in a multi-region edge internet network, characterized in that: The system comprises: A network access point, in response to receiving a task request, determines a service corresponding to the task request; obtains a scheduling policy corresponding to the service in a current scheduling period; determines a target area to which the task request is scheduled according to the scheduling policy; and sends the task request to a network access point corresponding to the target area; A central controller receives the scheduling cost accumulation, task arrival prediction information, resource allocation information and network status information corresponding to each service reported by each network access point in the multi-region edge interconnection network; the scheduling cost accumulation is the accumulation of scheduling cost of scheduling task requests for a service by the network access point in a scheduling cycle; at the end of the current scheduling cycle, for each service of each network access point, respectively, a computing power network graph is calculated according to the corresponding task arrival prediction information, resource allocation information and network status information; and a scheduling strategy for the next scheduling cycle is determined according to the scheduling cost accumulation and the computing power network graph; the scheduling strategy is the scheduling probability of scheduling the task request of the service to the edge computing resource pool of each region in the multi-region edge interconnection network; before the start of the next scheduling cycle, the scheduling strategy determined for each service of each network access point is issued to the corresponding network access point.

12. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the method according to any one of claims 1-4, or 5-8 is implemented.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method described in any one of claims 1 to 4 or 5 to 8 is implemented.

Images