Task scheduling and cache optimization method in storage and computation collaborative edge computing network

Abstract

The invention discloses a task scheduling and cache optimization method in a storage and calculation collaborative edge computing network, and aims to optimize the service delay cost of a user and the cache deployment cost of a service provider. The method comprises the following steps: modeling a reusable task scheduling and cache optimization problem in a storage-computing collaborative edge computing network, and decoupling the reusable task scheduling and cache optimization problem into two sub-problems: a task scheduling problem for minimizing service delay cost, and providing a dynamic task scheduling algorithm based on a merging mechanism for the task scheduling problem. The invention provides a calculation result caching algorithm combining dynamic planning and greedy, and further provides a forward-looking pre-caching strategy for solving the problem of minimizing the service caching cost. The task calculation scheduling and result caching method with reusability is provided for space-time correlation of task requests in a storage and calculation collaborative edge calculation network, strict performance guarantee is provided, the method has wide applicability, and the method is suitable for large-scale popularization and application. And challenges of resource limitation and high real-time requirements in a complex calculation scene can be efficiently dealt with.

Description

Technical Field

[0001] The present invention relates to the field of edge computing, and in particular to a method for task scheduling and cache optimization in a storage-computing collaborative edge computing network. Background Art

[0002] With the rapid development of artificial intelligence and Internet of Things technologies, computationally intensive and delay-sensitive mobile applications such as target detection, image recognition, augmented reality, and virtual reality experience games are constantly emerging. Generally speaking, achieving low latency requires ensuring computing speed while effectively controlling transmission delays. This requirement constitutes one of the main challenges facing traditional cloud computing architectures. Therefore, a new computing model is urgently needed to solve the above problems and improve the quality of service (QoS) for mobile users. Collaborative edge computing is considered to be a promising new computing paradigm. By splitting complex computing tasks into subtasks and offloading them to servers at the edge of the nearby network (such as 5G base stations, wireless access points, etc.), it can effectively reduce task processing delays and improve the utilization of computing resources. Compared with the unified management model of traditional cloud computing, collaborative edge computing is more flexible and efficient, and can better meet users' increasingly stringent service quality requirements.

[0003] Although collaborative edge computing has improved the utilization of computing resources in the network to a certain extent, computing redundancy still exists. The computing tasks requested by mobile users have certain correlations in many scenarios, including temporal correlation, spatial correlation, semantic correlation, etc. Repeated calculation of highly correlated request tasks will consume limited computing resources on the edge of the network, hindering the improvement of the service quality of edge computing. However, in recent years, most of the research work on scheduling optimization and service deployment optimization has ignored the correlation between computing tasks, resulting in the existence of computing redundancy. Therefore, it is necessary to combine the correlation of computing tasks in the resource allocation decision of edge computing.

[0004] Content Delivery Network (CDN) is a new type of network infrastructure generated by the rapid development of social media today. CDN reduces the access latency of user requests by storing specified content in the streaming media resource library on servers closer to users. The caching technology combined with CDN in edge computing has received widespread attention from scholars, but it is limited to caching the service data required for computing (such as execution programs, containers, databases, etc.) to the edge server, and does not involve caching the computing results. Since the correlation of the above computing tasks is common in many scenarios, caching and reusing the computing results can reduce computing redundancy and further improve computing efficiency.

[0005] In summary, the present invention combines the relevance of computing tasks and the cache reuse of computing results to design a task scheduling and cache optimization method in a storage-computing collaborative edge computing network. Summary of the invention

[0006] The purpose of the present invention is to provide a method for task scheduling and cache optimization in a storage-computing collaborative edge computing network, which is aimed at resource-constrained edge computing networks to optimize the user's delay cost and the supplier's cache cost. First, for the problem of related task scheduling in a storage-computing collaborative edge computing network, a storage-computing collaborative architecture is proposed and system modeling is performed, and the optimization goal is to minimize the user's delay cost and the supplier's cache cost. Secondly, the problem of related task scheduling in a storage-computing collaborative edge computing network is decomposed into two sub-problems and two corresponding methods and a pre-caching mechanism are proposed to construct a complete and reusable task scheduling and caching strategy.

[0007] To achieve the above functions, the present invention designs a task scheduling and cache optimization method in a storage-computing collaborative edge computing network, and executes the following steps S1-S2 to minimize the user's delay cost and the service provider's cache cost:

[0008] Step S1: System modeling is performed for the reusable task scheduling and cache optimization problems in the storage-computing collaborative edge computing network, and the overall system model of the storage-computing collaborative edge computing network, the storage-computing collaborative edge computing network architecture, the computing request task offloading model, and the reusable computing result data block cache model are constructed. The optimization goal of the model is defined as minimizing the user's delay cost and the service provider's cache cost;

[0009] Step S2: Decompose the reusable task scheduling and cache optimization problem in the storage-computing collaborative edge computing network into two sub-problems and solve them separately, including the following steps:

[0010] Step S2.1: The first sub-problem is to determine the scheduling scheme for each computing request task and its subtasks to optimize the user's delay cost; in view of the time-space correlation characteristics of computing request tasks, a dynamic task scheduling algorithm based on the merging mechanism is proposed. First, the computing request tasks are dynamically sorted and then scheduled; at the same time, repeated computing request tasks are merged, and finally the execution order of the data blocks to be processed is further optimized based on the idea of ​​box packing;

[0011] Step S2.2: The second sub-problem is to determine the data result caching strategy on each server, optimize the caching cost of the service provider, and propose an approximate algorithm that combines dynamic programming and greedy to optimize the caching strategy for each edge server; in view of the dynamic changes in the timeliness of cached data results, a pre-caching mechanism is proposed to further optimize the caching cost.

[0012] The present invention also designs a computer device, including a memory, a processor and a computer program stored in the memory, wherein the processor executes the computer program to implement the task scheduling and cache optimization method in a storage-computing collaborative edge computing network.

[0013] The present invention also designs a computer-readable storage medium on which a computer program / instruction is stored. When the computer program / instruction is executed by a processor, the method for task scheduling and cache optimization in a storage-computing collaborative edge computing network is implemented.

[0014] The present invention also designs a computer program product, including a computer program / instruction, which, when executed by a processor, implements the task scheduling and cache optimization method in a storage-computing collaborative edge computing network.

[0015] Beneficial effects: Compared with the prior art, the advantages of the present invention include:

[0016] The method of the present invention decomposes the problem of joint scheduling and cache optimization of reusable tasks in the storage-computing collaborative edge computing network into two sub-problems, decoupling the decision variables while reducing the complexity of the problem;

[0017] First, the first sub-problem takes minimizing the user's delay cost as the optimization goal. In this stage, a dynamic task scheduling algorithm based on the merging mechanism is proposed, which can be proven to be The approximation ratio of

[0018] The second sub-problem determines the data result caching strategy on each server, optimizes the cache cost of the service provider, and proposes an approximate algorithm combining dynamic programming and greedy, which can be proven to be The approximation ratio of

[0019] In addition, in order to make full use of computing resources, a forward-looking pre-caching strategy is proposed. In summary, the present invention decouples the complex problem into two sub-problems through an innovative method, optimizes the user's delay cost and the service provider's cache cost, and provides a strict performance guarantee for the proposed algorithm, ensuring the feasibility of our method. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of an overall system model of a storage-computing collaborative edge computing network provided according to an embodiment of the present invention;

[0021] Figure 2 is a flow chart of a first-stage computing request task scheduling algorithm provided according to an embodiment of the present invention;

[0022] Figure 3 It is a schematic diagram of second-stage cache optimization provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0023] The present invention will be further described below in conjunction with the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the protection scope of the present invention.

[0024] An embodiment of the present invention provides a method for task scheduling and cache optimization in a storage-computing collaborative edge computing network, which performs the following steps S1-S2 to minimize the user's delay cost and the service provider's cache cost:

[0025] Step S1: System modeling is performed for the reusable task scheduling and cache optimization problems in the storage-computing collaborative edge computing network, and the overall system model of the storage-computing collaborative edge computing network, the storage-computing collaborative edge computing network architecture, the computing request task offloading model, and the reusable computing result data block cache model are constructed. The optimization goal of the model is defined as minimizing the user's delay cost and the service provider's cache cost;

[0026] The specific steps of step S1 are as follows:

[0027] Step S1.1: Construct the overall system model of the storage-computing collaborative edge computing network; the storage-computing collaborative edge computing network includes N base stations (Base Station, BS) and an SDN controller (Software-defined Networking Controller). The SDN-based network architecture separates the control plane from the data plane. Each base station is equipped with an edge server to form an edge node. All edge nodes are connected by wired links. Represents the set of edge nodes; considering the heterogeneity of edge nodes, C n and S n To represent the computing power and storage capacity of edge nodes respectively, the time in the network system is divided into T time slots, and the time slot set is Each time slot The length of is τ. In each time slot, a user device generates at most one computing request and sends it to the nearest edge node through the wireless network. The input data required by the computing request is provided by the relevant sensors. Different sensors transmit data to the nearest server in each time slot. Depending on the scenario, the sensor can be located at the user's own location or at a different location from the user. Sensor data is time-sensitive and will only be calculated when and only when requested by the user. Otherwise, it will be automatically discarded after expiration. The computing request task set received by the edge node n at time t is used Indicates that the set of computing request tasks received by all edge nodes is recorded as The computing request task type is represented by θ (θ∈Θ); with the help of the SDN controller, the storage location of all cached computing results is accessible to all edge nodes, and all user requests are summarized at the SDN controller, and the computing request task offloading scheduling is performed;

[0028] like Figure 1 As shown, the storage and computing collaborative edge computing network in this embodiment is divided into three layers: data collection layer, request service layer, and scheduling control layer. The data collection layer is a variety of sensors that are responsible for providing users with raw data and sending it regularly to the nearest edge node for calculation. The request service layer includes user devices that generate computing tasks and collaborative edge servers that perform computing tasks. The original data is scheduled to different edge nodes for calculation only after being requested by the user, otherwise it will be automatically deleted after the data expires. The edge server has the ability to calculate and cache data results. If the SDN controller queries that the collaborative edge node has cached the data results requested by the user, there is no need to recalculate. The scheduling control layer includes the SDN controller of the entire network, which is responsible for scheduling user requests and querying the cache status of each edge node.

[0029] Step S1.2: Construct a storage-computing collaborative edge computing network architecture; the input data of the computing request task is often divisible (such as pictures, 3D point cloud frames, etc.), so the computing request task can be partially offloaded to different edge nodes for calculation to improve efficiency. All possible input data sets of each computing request task can be divided into a series of data blocks, and the division strength of different computing request tasks is determined by different computing request task types. Suppose the input data set of the computing request task of type θ is divided into Divide all possible input data sets of each computing request task into multiple data blocks, and the size of each data block is expressed as The kth computing request task type received by edge node n is θ, and its input data is represented as a binary vector If the user's calculation requests an input data block Then the corresponding element otherwise, Therefore, the size of all input data blocks of the computing request task k of type θ received by the edge node n is expressed as: Will satisfy The set of all data blocks is defined as represents the actual data block set of computing request task k; the offloading decision variable of computing request task k accepted by edge node n is represented by J θ ×N-dimensional vector representation: in, is an N-dimensional vector, that is when , it means that the j-th data block of computing request task k is unloaded from edge node n to edge node m; otherwise

[0030] Step S1.3: Construct a computing request task offloading model; wireless links are used for data transmission between user devices and edge nodes. Wireless channel resources are shared by multiple user devices. The rate at which computing request tasks k are transmitted to edge nodes is in represents the interference caused by other simultaneous transmission devices, B is the wireless channel bandwidth, is the transmission power, represents the channel gain, N 0 represents the Gaussian noise power spectral density, so the calculation request task transmission delay is expressed as Where S k is the size of the user's request data packet; the wired link transmission delay of the jth data block of the request task k is expressed as in represents the wired transmission rate between edge nodes n and m; edge nodes Allocate computing resources according to the computing request task type. Therefore, the computational delay is expressed as where μ k represents the computing density; after the computing request task is unloaded to the edge node, it will enter the computing request task queue of this type; the queuing delay of the jth data block of the computing request task k is in Q(j′) is the indicator function, and the symbol ≤ indicates the priority order. represents the priority of data block j in task k, represents the priority of data block j′ of task k′, Q(j′)=1 means that data block j can only be calculated after data block j′ is processed; the retrieval delay is the time for the edge node to retrieve whether the calculation result of the calculation request task is cached from the SDN controller, which is expressed as where f r Indicates that the retrieval time is a function of the number of data to be retrieved; if the calculation result of a data block of the calculation request task has been cached, only the calculation result needs to be transmitted instead of recalculated; The calculation result of the data block j is cached on the edge node m; the merging delay is the time to merge and reorganize the results of all data blocks, expressed as where f mIt means that the merging time is a function of the number of data blocks to be merged. In different tasks, the scale of input data and output data may be significantly different. In the three-dimensional target detection task where the amount of input data is much larger than the amount of output data, the input data is a large amount of point cloud data, and the output is only the target bounding box information. On the contrary, in the museum AR rendering task, the input data is the location information of the visitor, and the output is the rendering data corresponding to the location. The output data volume far exceeds the input. Without loss of generality, the present invention only considers one-way (large-scale data) transmission delay. The total delay of the calculation request task k (type θ) is:

[0031]

[0032] From the user's perspective, when the actual delay of a task exceeds the user's maximum tolerable delay, the user's dissatisfaction will increase. Therefore, the user's delay cost is:

[0033]

[0034] in is the maximum delay that the user can tolerate, w k The price to be paid for exceeding the maximum tolerable delay;

[0035] Step S1.4: Construct a reusable computing result data block cache model; the i-th reusable result data block cached by edge node n is represented as Among them, I i Indicates the calculation input data or input data index, O i represents the calculation result, θ i Indicates the type of computing request task, V i Indicates the value of data. Indicates the validity period of the data. The value of the calculation result that exceeds the validity period is 0. The data to be cached on each edge node includes new data and old data. Since the edge servers are collaborative, the data blocks of the calculation request task may be unloaded to different servers for calculation. This part of data is called new data. The old data is the calculation result data block that has been stored on the server. The local cache hit rate of data is obtained from historical information, which represents the probability of requesting a certain cached data among all requests received by edge node n. The local cache hit rate of data is expressed as in is the number of times edge node n utilizes reusable data i in all computation requests received in time slot t; Age of Data (AoD) represents the freshness of data on a time scale, denoted as A i(t) = tg(t), where g(t) is the timestamp when the data is cached; the higher the freshness of the data, the greater the probability that it may be reused in the future. Represents, where λ is an adjustable parameter, represents the maximum validity period of data block i. The data value of the old data block is expressed as The data value of the new data block is expressed as in is the average wired transmission rate from edge node n to other edge nodes, is the average local cache hit rate in the system, γ j Represents the data age parameter; the cache decision consists of two parts: whether the calculation result of the newly arrived request needs to be cached, and Indicates whether the cached calculation result data is retained. express;

[0036] From the supplier's perspective, cache data must bear the cost of using storage hardware. Storage resources are fixed resources, and their fixed cost is defined as C fix ; Only by caching valuable data can user satisfaction be improved and redundant computing be reduced. The caching cost of the service provider C cache It is expressed as:

[0037]

[0038] Step S1.5: Define the optimization objective of the model as minimizing the user's delay cost and the service provider's caching cost.

[0039] In step S1.5, the delay cost of the user and the cache cost of the service provider are minimized as follows:

[0040] The input of the model is defined as: the set of computing request tasks received by all edge nodes The set of computing request tasks received by edge node n Edge node collection The set of data blocks that the computing request task k requests to compute The computing power C of edge node n n and storage capacity S n , calculate the set of all data blocks I with the requested task type θ θ , and calculate the data block size Z of the request task type θ θ ;

[0041] The output of the model is defined as: the unloading decision for each data block Decision making on the order of data block computations offloaded to the same server Caching decisions for each data block

[0042] The optimization objectives of the model are as follows:

[0043]

[0044] st

[0045]

[0046] Among them, C1 ensures that the data blocks of the task are completely offloaded to the server for calculation, C2 ensures that each offloaded data block can be processed by at most one edge node, C3 and C4 represent cache decisions, indicating whether a data block should be retained or discarded, C5 represents the limitation of storage capacity, and C6 represents the limitation of computing capacity.

[0047] Step S2: Decompose the reusable task scheduling and cache optimization problem in the storage-computing collaborative edge computing network into two sub-problems and solve them separately, including the following steps:

[0048] Step S2.1: The first sub-problem is to determine the scheduling scheme for each computing request task and its sub-tasks to optimize the user's delay cost; when the actual task delay exceeds the user's expected maximum tolerable delay, it will lead to a decline in user experience, thereby incurring corresponding costs. The first sub-problem is stated as: in, In view of the spatiotemporal correlation characteristics of computing request tasks, a dynamic task scheduling algorithm based on a merging mechanism is proposed, and the optimal solution satisfies the following two properties. Property 1: There exists an optimal scheduling, and all the data blocks to be calculated that are scheduled to the same task on the same server will be processed continuously. Property 2: There exists an optimal scheduling, and the arrangement order of all data blocks processed on all servers is the same. Therefore, it is necessary to determine the scheduling order of tasks and the scheduling order of subtasks. First, the computing request tasks are dynamically sorted and then scheduled; at the same time, the repeated computing request tasks are merged, and finally the execution order of the data blocks to be processed is further optimized based on the idea of ​​box packing; the method proposed in the present invention has The flow chart of the whole algorithm is as follows: Figure 2 shown.

[0049] Step S2.1 includes the following steps:

[0050] Step S2.1.1: Dynamically sort all computing request tasks; first sort the execution time of all data blocks of each computing request task in descending order, and schedule all data blocks of the computing request task to the first idle server to calculate the completion time T of each computing request task. k ; Find T kThe smallest computing request task k' is scheduled after the data blocks of the computing request task are merged. After the computing request task k' is scheduled, the above process is repeated until all computing request tasks are scheduled. It is worth noting that since the load of the server changes with the scheduling order, the sorting process is dynamic.

[0051] Step S2.1.2: Merge the data blocks to be processed of different computing request tasks; first count the number of times each data block is requested, and form a new set Ω for the data blocks with a number greater than 1; according to the result of dynamic sorting of computing request tasks, after computing request task k' is executed, subsequent computing request tasks do not need to repeatedly calculate the results of reusable data blocks, so update Ω to

[0052] Step S2.1.3: Schedule all pending data blocks of task k'. Since the initial completion time T of task k' has been obtained k′ , T k′ Set to "box size". Use the idea of ​​box packing to rearrange the data blocks and further optimize the load, that is, within T k′ In the case of , the data block with the longest processing time is scheduled to the server with the heaviest load as much as possible. If scheduling the data block to the server with the heaviest load still violates T k′ If the limit is exceeded, it will be scheduled to the server with the second largest load for calculation. And so on, if the data block is assigned to the server with the smallest load and still exceeds T k′ , it will be assigned to this server and T k′ Update to the completion time of the task. Repeat the above process until all data blocks are scheduled.

[0053] Step S2.2: The second sub-problem determines the data result caching strategy on each server and optimizes the service provider's caching cost, that is, the difference between the storage hardware fee paid by the provider for caching data and the caching benefit. Since the offloading decision has been solved by the first sub-problem, the cost minimization problem can be converted into a value benefit maximization problem, that is, To solve this problem, a dynamic programming and greedy approximation algorithm is proposed to optimize the cache strategy for each edge server. The approximation ratio is guaranteed. In view of the dynamic changes in the timeliness of cached data results, a pre-caching mechanism is proposed to further optimize the cache cost; the solution strategy diagram of the second sub-problem is shown in Figure 3 shown.

[0054] Step S2.2 includes the following steps:

[0055] Step S2.2.1: Define the set of calculation results of the data blocks to be cached (hereinafter referred to as data blocks) as For each data block in U(t), renumber the index and simplify the expression to ensure in is the value of data block i, s i is the size of the data block; in order to balance the high time overhead of the dynamic programming algorithm and the insufficient performance of the greedy algorithm, the data blocks are divided into two sets according to the value of the data: the high-value data block set and a collection of low-value data blocks If satisfied Then data block i belongs to Otherwise in And meet right The data blocks in the value are scaled, that is, in ∈ is an adjustable parameter; scaling a part of the data blocks is to further reduce the problem size, reduce time complexity, and ensure a good approximation ratio.

[0056] Step S2.2.2: For the data blocks in the high-value data block set, a dynamic programming algorithm is used; the value maximization problem is transformed into a problem of minimizing the occupied space under the condition of achieving a certain value. G(i,V) is used to represent the minimum total space consumed when selecting from the first i data blocks to make the total value V. Therefore It is worth noting that the value here refers to the scaled value. Through the dynamic programming method, the total value can be obtained from 0 to The set of data blocks selected at the time. In other words, each solution obtained by dynamic programming is selected in turn. The data blocks in are combined until the solution with the highest value is found.

[0057] Step S2.2.3: Since the validity of cached data results changes over time, relying solely on the above algorithm to make caching decisions may have certain fluctuations. The edge server may cache some data blocks of low value when a large number of data blocks expire. Therefore, a sampling-based random pre-caching strategy is proposed to calculate the original data that may be reused in the future in advance when the server is idle, and cache it in advance to prevent batch expiration of data. This strategy first determines whether there is still idle time before the next time slot when the server calculates all tasks in this batch. If there is idle time, a data block is randomly selected from the cache list, and the probability of P is used to decide whether to recalculate the original data and replace the current data block. Probability Where T maxis the time point when the data block expires, and w is an adjustable parameter. The closer the data block is to the expiration point, the more likely it is to be selected for re-precomputation with a high probability. represents the future benefit if data block i is pre-cached, where Z i represents the size of data block i, P represents the probability that the data block will be requested in the future, T valid Indicates the validity period of the data block; L * Recalculate data block i * The future benefits after that, if L * >L then L * Replace it with a new round of L; after n rounds of iteration, select the data block that meets the requirements and recalculate to complete the pre-caching.

[0058] An embodiment of the present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the task scheduling and cache optimization method in a storage-computing collaborative edge computing network.

[0059] An embodiment of the present invention also provides a computer-readable storage medium having a computer program / instruction stored thereon. When the computer program / instruction is executed by a processor, the method for task scheduling and cache optimization in a storage-computing collaborative edge computing network is implemented.

[0060] An embodiment of the present invention also provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the method for task scheduling and cache optimization in a storage-computing collaborative edge computing network.

[0061] The embodiments of the present invention are described in detail above with reference to the accompanying drawings, but the present invention is not limited to the above embodiments, and various changes can be made within the knowledge scope of ordinary technicians in this field without departing from the purpose of the present invention.

Claims

1. A method for task scheduling and cache optimization in a storage-computing collaborative edge computing network, characterized in that: Execute the following steps S1-S2 to minimize the user's delay cost and the service provider's cache cost: Step S1: System modeling is performed for the reusable task scheduling and cache optimization problems in the storage-computing collaborative edge computing network, and the overall system model of the storage-computing collaborative edge computing network, the storage-computing collaborative edge computing network architecture, the computing request task offloading model, and the reusable computing result data block cache model are constructed. The optimization goal of the model is defined as minimizing the user's delay cost and the service provider's cache cost; Step S2: Decompose the reusable task scheduling and cache optimization problem in the storage-computing collaborative edge computing network into two sub-problems and solve them separately, including the following steps: Step S2.1: The first sub-problem is to determine the scheduling scheme for each computing request task and its subtasks to optimize the user's delay cost; in view of the time-space correlation characteristics of computing request tasks, a dynamic task scheduling algorithm based on the merging mechanism is proposed. First, the computing request tasks are dynamically sorted and then scheduled; at the same time, repeated computing request tasks are merged, and finally the execution order of the data blocks to be processed is further optimized based on the idea of ​​box packing; Step S2.2: The second sub-problem is to determine the data result caching strategy on each server, optimize the caching cost of the service provider, and propose an approximate algorithm that combines dynamic programming and greedy to optimize the caching strategy for each edge server; in view of the dynamic changes in the timeliness of cached data results, a pre-caching mechanism is proposed to further optimize the caching cost.

2. According to the method of claim 1, the task scheduling and cache optimization method in the storage-computing collaborative edge computing network is characterized in that: The specific steps of step S1 are as follows: Step S1.1: Construct the overall system model of the storage-computing collaborative edge computing network; the storage-computing collaborative edge computing network includes N base stations and an SDN controller. Each base station is equipped with an edge server to form an edge node. All edge nodes are connected by wired links. Represents the set of edge nodes; considering the heterogeneity of edge nodes, C n and S n To represent the computing power and storage capacity of edge nodes respectively, the time in the network system is divided into T time slots, and the time slot set is Each time slot The length of is τ. In each time slot, a user device generates at most one computing request and sends it to the nearest edge node through the wireless network. The input data required by the computing request is provided by the relevant sensors. Different sensors transmit data to the nearest server in each time slot. The sensor data is time-sensitive and will be calculated only when and only when requested by the user. Otherwise, it will be automatically discarded after expiration. The computing request task set received by the edge node n at time t is Indicates that the set of computing request tasks received by all edge nodes is recorded as The computing request task type is represented by θ (θ∈Θ); With the help of SDN controller, the storage location of all cached computation results is accessible to all edge nodes, and all user requests are aggregated at the SDN controller, where computation request tasks are offloaded and scheduled; Step S1.2: Construct a storage-computing collaborative edge computing network architecture; suppose the input data set of the computing request task of type θ is divided into Divide all possible input data sets of each computing request task into multiple data blocks, and the size of each data block is expressed as The kth computing request task type received by edge node n is θ, and its input data is represented as a binary vector If the user's calculation requests an input data block Then the corresponding element otherwise, Therefore, the size of all input data blocks of the computing request task k of type θ received by the edge node n is expressed as: Will satisfy The set of all data blocks is defined as represents the actual data block set of computing request task k; the offloading decision variable of computing request task k accepted by edge node n is represented by J θ ×N-dimensional vector representation: in, is an N-dimensional vector, that is when , it means that the j-th data block of computing request task k is unloaded from edge node n to edge node m; otherwise Step S1.3: Construct a computing request task offloading model; wireless links are used for data transmission between user devices and edge nodes. Wireless channel resources are shared by multiple user devices. The rate at which computing request tasks k are transmitted to edge nodes is in represents the interference caused by other simultaneous transmission devices, B is the wireless channel bandwidth, is the transmission power, represents the channel gain, and N0 represents the Gaussian noise power spectrum density. Therefore, the transmission delay of the calculation request task is expressed as Where S k is the size of the user's request data packet; the wired link transmission delay of the jth data block of the request task k is expressed as in represents the wired transmission rate between edge nodes n and m; edge nodes Allocate computing resources according to the computing request task type. Therefore, the computational delay is expressed as where μ k represents the computing density; after the computing request task is unloaded to the edge node, it will enter the computing request task queue of this type; the queuing delay of the jth data block of the computing request task k is in in represents the priority of data block j in task k, represents the priority of data block j' of task k'. Q(j') is an indicator function, the symbol ≤ represents the priority order, Q(j') = 1 means that data block j needs to wait until data block j' is processed before it can be calculated; the retrieval delay is the time it takes for the edge node to retrieve from the SDN controller whether the calculation result of the calculation request task is cached, expressed as where f r Indicates that the retrieval time is a function of the number of data to be retrieved; The calculation result of the data block j is cached on the edge node m; the merging delay is the time to merge and reorganize the results of all data blocks, expressed as where f m It indicates that the merging time is a function of the number of data blocks to be merged. The total delay of the request task k is calculated as: The delay cost for the user is: in is the maximum delay that the user can tolerate, w k The price to be paid for exceeding the maximum tolerable delay; Step S1.4: Construct a reusable computing result data block cache model; the i-th reusable result data block cached by edge node n is represented as Among them, I i Indicates the calculation input data or input data index, O i represents the calculation result, θ i Indicates the type of computing request task, V i Indicates the value of data. Indicates the validity period of the data. The value of the calculation result that exceeds the validity period is 0. The data to be cached on each edge node includes new data and old data. Since the edge servers are collaborative, the data blocks of the calculation request task may be unloaded to different servers for calculation. This part of data is called new data. The old data is the calculation result data block that has been stored on the server. The local cache hit rate of data is expressed as in is the number of times edge node n utilizes reusable data i in all computation requests received in time slot t; data age represents the freshness of data on a time scale, denoted as A i (t) = tg(t), where g(t) is the timestamp when the data is cached; the data age parameter is Represents, where λ is an adjustable parameter, represents the maximum validity period of data block i. The data value of the old data block is expressed as The data value of the new data block is expressed as where γ j Indicates the data age of the data block. is the average wired transmission rate from edge node n to other edge nodes, is the average local cache hit rate in the system; the cache decision includes two parts: whether the calculation result of the newly arrived request needs to be cached, and Indicates whether the cached calculation result data is retained. express; The service provider’s caching cost C cache It is expressed as: In the formula, Cfix is ​​the fixed cost; Step S1.5: Define the optimization objective of the model as minimizing the user's delay cost and the service provider's caching cost.

3. According to claim 2, a method for task scheduling and cache optimization in a storage-computing collaborative edge computing network is characterized in that: In step S1.5, the delay cost of the user and the cache cost of the service provider are minimized as follows: The input of the model is defined as: the set of computing request tasks received by all edge nodes The set of computing request tasks received by edge node n Edge node collection The set of data blocks that the computing request task k requests to compute The computing power C of edge node n n and storage capacity S n , calculate the set of all data blocks I with the requested task type θ θ , and calculate the data block size Z of the request task type θ θ ; The output of the model is defined as: the unloading decision for each data block Decision making on the order of data block computations offloaded to the same server Caching decisions for each data block The optimization objectives of the model are as follows: st Among them, C1 ensures that the data blocks of the task are completely offloaded to the server for calculation, C2 ensures that each offloaded data block can be processed by at most one edge node, C3 and C4 represent cache decisions, indicating whether a data block should be retained or discarded, C5 represents the limitation of storage capacity, and C6 represents the limitation of computing capacity.

4. According to the method of claim 1, the task scheduling and cache optimization method in the storage-computing collaborative edge computing network is characterized in that: Step S2.1 includes the following steps: Step S2.1.1: Dynamically sort all computing request tasks; first sort the execution time of all data blocks of each computing request task in descending order, and schedule all data blocks of the computing request task to the first idle server to calculate the completion time T of each computing request task. k ; Find T k The smallest computing request task k' is scheduled after the data blocks of the computing request task are merged. After the computing request task k' is scheduled, the above process is repeated until all computing request tasks are scheduled. Step S2.1.2: Merge the data blocks to be processed of different computing request tasks; first count the number of times each data block is requested, and form a new set Ω for the data blocks with a number greater than 1; according to the result of dynamic sorting of computing request tasks, after computing request task k' is executed, subsequent computing request tasks do not need to repeatedly calculate the results of reusable data blocks, so update Ω to Step S2.1.3: Schedule all pending data blocks of computing request task k'; within the completion time T of computing request task k' k′ In the case of T, the data block with long processing time is scheduled to the server with the largest load as much as possible; if scheduling the data block to the server with the largest load still violates T k′ If the limit is exceeded, it will be scheduled to the server with the second largest load for calculation; and so on. If the data block is assigned to the server with the smallest load and still exceeds T k′ , it will be assigned to this server and T k′ Update; repeat the above process until all data blocks are scheduled.

5. According to claim 4, a method for task scheduling and cache optimization in a storage-computing collaborative edge computing network is characterized in that: Step S2.2 includes the following steps: Step S2.2.1: Define the calculation result set of the data block to be cached as For each data block sort in U(t), ensure in is the value of data block i, s i is the size of the data block; the data blocks are divided into two sets according to the value of the data: high-value data block set and a collection of low-value data blocks If satisfied Then data block i belongs to Otherwise in And meet right The data blocks in the value are scaled, that is, in ∈ is an adjustable parameter; Step S2.2.2: For the data blocks in the high-value data block set, a dynamic programming algorithm is used; the value maximization problem is transformed into a problem of minimizing the occupied space under the condition of achieving a certain value. G(i,V) is used to represent the minimum total space consumed when selecting from the first i data blocks to make the total value V. Therefore Get the total value from 0 to The set of data blocks selected when the dynamic programming is performed is combined with Combine the data blocks in until the solution with the highest value is found; Step S2.2.3: Determine whether there is still idle time before the next time slot when the server has completed all the calculation request tasks of this batch. If there is idle time, randomly select a data block in the cache list and decide whether to recalculate the original data block with a probability of P, and replace the current data block; probability Where T max is the time point when the data block expires, and w is an adjustable parameter; represents the future benefit if data block i is pre-cached, where P represents the probability that the data block will be requested in the future, and Z i represents the size of data block i, T valid Indicates the validity period of the data block; L * Recalculate data block i * The future benefits after that, if L * >L then L * Replace it with a new round of L; after n rounds of iteration, select the data block that meets the requirements and recalculate to complete the pre-caching.

6. A computer device comprising a memory, a processor and a computer program stored in the memory, characterized in that: The processor executes the computer program to implement a method for task scheduling and cache optimization in a storage-computing collaborative edge computing network as described in any one of claims 1-5.

7. A computer-readable storage medium having a computer program / instruction stored thereon, characterized in that: When the computer program / instruction is executed by the processor, it implements the method for task scheduling and cache optimization in a storage-computing collaborative edge computing network as described in any one of claims 1-5.

8. A computer program product comprising a computer program / instructions, characterized in that When the computer program / instruction is executed by the processor, it implements the method for task scheduling and cache optimization in a storage-computing collaborative edge computing network as described in any one of claims 1-5.

Images