Chain network convergence architecture resource scheduling method and system oriented to maximization of computational efficiency

Abstract

The invention provides a computing efficiency maximization-oriented chain network convergence architecture resource scheduling method, which comprises the following steps of: acquiring full-path structure information of all to-be-scheduled DAG (Directed Acyclic Graph) tasks in a system, including a subtask list and a dependency relationship of each path; calculating the residual calculation amount of each path, and selecting the path with the maximum residual calculation amount as a key path; calculating the urgency value of each DAG task based on the residual calculation amount and the residual time of the key path; selecting a first schedulable sub-task of which all the front tasks are completed from the critical path with the highest urgency value; the expected completion time of the schedulable sub-task on each candidate server is evaluated; selecting a server capable of completing the task earliest as a scheduling target; and repeating the step 1 to the step 5 until all the subtasks of all the DAG tasks are scheduled. According to the method, joint modeling of task urgency and equipment service capability is introduced, so that the reasonability of resource scheduling and task completion efficiency are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of blockchain, in particular to a chain network fusion architecture resource scheduling method and system for maximizing computing efficiency. BACKGROUND

[0002] In recent years, with the rapid development of edge computing, blockchain and Internet of Things (IoT) technologies, distributed scheduling of computing tasks has become a key link to improve the overall performance of the system. In particular, in the chain network fusion architecture, computing resources exhibit strong heterogeneity and large fluctuations in availability, and the rationality of the resource scheduling strategy directly affects the computing efficiency and task completion timeliness of the system.

[0003] Existing resource scheduling methods have achieved certain results in technical implementation, but still have the following limitations: on the one hand, most scheduling methods assume that the task model is static and single-task, ignoring the fact that tasks generated by IoT devices generally have dependency relationships, making it difficult to adapt to the widely existing DAG (Directed Acyclic Graph) task structure in reality; on the other hand, scheduling strategies rely on simple heuristic algorithms (such as shortest execution time first, greedy algorithm, etc.), which cannot guarantee optimal computing efficiency in the case of large task size or device performance heterogeneity. In addition, in recent years, some scheduling methods based on neural networks or reinforcement learning have emerged, although they have achieved good results in specific scenarios, but the model training relies on a large amount of data and lacks adaptability to changes in system topology, making it difficult to generalize to a general scheduling environment. SUMMARY

[0004] To solve the above problems, a chain network fusion architecture resource scheduling method for maximizing computing efficiency is proposed. This method takes DAG tasks as the modeling object, designs a sub-task scheduling priority evaluation method based on the critical path urgency, and selects the most suitable server for task allocation based on the principle of maximizing scheduling benefit, and introduces the joint modeling of task urgency and device service capability to improve the rationality of resource scheduling and task completion efficiency.

[0005] To solve the above technical problems, the present application adopts the following technical solutions: A chain network fusion architecture resource scheduling method for maximizing computing efficiency, comprising the following steps: Step 1: Obtain the full path structure information of all DAG tasks to be scheduled in the system, including the sub-task list and dependency relationship of each path; Step 2: Calculate the remaining computing amount of each path, and select the path with the maximum remaining computing amount as the critical path; Step 3: Calculate the urgency value of each DAG task based on the critical path remaining computing amount and remaining time; Step 4: Select the first schedulable subtask from the critical path with the highest urgency value, where all preceding tasks have been completed; and evaluate the expected completion time of the schedulable subtask on each candidate server; Step 5: Select the server that can complete the task earliest as the scheduling target; Step 6: Repeat steps 1 to 5 until all subtasks of all DAG tasks have been scheduled.

[0006] Furthermore, the formula for calculating the remaining computational cost for each path in step 2 is as follows:

[0007] in, Represents path P i The remaining computational load, Representing a path Neutron mission The computational load, This represents the set of subtasks that have not yet been scheduled.

[0008] Furthermore, the formula for calculating the urgency value in step 3 is as follows:

[0009] in, This represents the urgency value for each DAG task; This represents the remaining computational cost of the critical path. This represents the remaining time until the task's deadline.

[0010] Furthermore, the remaining time until the task's deadline. The calculation formula is as follows:

[0011] in, This indicates the latest allowed completion time for the DAG task. This indicates the current scheduling time.

[0012] Furthermore, the selection rules for the schedulable subtasks are as follows: On the critical path with the highest urgency value, select the first subtask where all dependent tasks have been completed but not yet executed.

[0013] Furthermore, the formula for calculating the expected completion time of the schedulable subtask on each candidate server in step 4 is as follows:

[0014] in, The expected completion time of the schedulable subtask on each candidate server; representing schedulable sub-tasks latency on candidate servers latency on candidate servers representing schedulable sub-tasks execution time on candidate servers execution time on candidate servers

[0015] Further, the step 5 comprises: comparing the expected completion time of the schedulable sub-tasks on each server selecting the server with the smallest expected completion time as the scheduling target.

[0016] Further, the step 6 further comprises: updating the following states in real time after each schedulable sub-task is scheduled: the set of schedulable sub-tasks in the DAG path; the remaining computation amount of each path Li and the critical path; the waiting queue state of each server.

[0017] In another aspect, the present application provides a chain-network fusion architecture resource scheduling system for maximizing computing efficiency, comprising: a path analysis module: used for obtaining the full path structure information of all schedulable DAG tasks in the system, including the sub-task list and dependency relationship of each path; a critical path calculation module: used for calculating the remaining computation amount of each path, and selecting the path with the largest remaining computation amount as the critical path; an urgency evaluation module: used for calculating the urgency value of each DAG task based on the remaining computation amount and the remaining time of the critical path; a scheduling decision module: used for selecting the first schedulable sub-task whose all preceding tasks have been completed from the critical path with the highest urgency value; and evaluating the expected completion time of the schedulable sub-task on each candidate server; a scheduling target determination module: used for selecting the server that can complete the task earliest as the scheduling target; a state updating module: used for updating the task state and the server resource state in real time.

[0018] Further, the execution logic of the scheduling decision module comprises: sorting the schedulable DAG tasks based on the urgency value U selecting the first schedulable sub-task on the critical path of the selected DAG; generating an expected completion time sorting list of the schedulable sub-task on each candidate server according to the server latency and computing capacity; assigning the sub-task to the server with the smallest expected completion time. ​

[0019] Compared with the prior art, the present application has the following beneficial effects: 1. The present application introduces the structural features and urgency of DAG tasks into the scheduling strategy, improving the task priority discrimination ability.

[0020] 2. The present application introduces server service capability, combines actual waiting time and computing power to allocate resources.

[0021] 3. The present application minimizes the completion time of subtasks, improves the overall computing efficiency of the system and the completion time of DAG tasks.

[0022] 4. The method of the present application can adapt to various chain network fusion scenarios and has universality and expansibility. BRIEF DESCRIPTION OF DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0024] Figure 1 is a method execution flowchart of the present application.

[0025] Figure 2 is a DAG task static, dynamic to DAG diagram of the present application. DETAILED DESCRIPTION

[0026] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below. In the following description, a lot of specific details are set forth in order to fully understand the present application. However, the present application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar improvements without departing from the connotation of the present application, therefore the present application is not limited by the specific embodiments disclosed below. Example 1 The embodiment of the application adopts Python language and networkx library, proposes and implements a networking fusion architecture resource scheduling method for maximizing computing efficiency, aims to select the most suitable server for task allocation based on critical path urgency and scheduling benefit maximization principle, so as to support dynamic and static multi-DAG task scheduling scenarios in chain-network fusion architecture. In the embodiment, a simple scheduling scenario containing two DAG tasks is used. The first DAG is denoted as G1, containing four nodes {A, B, C, D}, and the path information is {(A, B), (B, C), (C, D), (A, D)}, and the computing amount of the subtasks represented by the nodes is WA, WB, WC, and WD respectively. The second DAG is denoted as G2, containing four nodes {E, F, G, H}, and the path information is {(E, H), (E, F), (F, G), (G, H)}, and the computing amount of the subtasks represented by the nodes is WE, WF, WG, and WH respectively. E , W F , W G , W H .

[0027] For details Figure 1 , the embodiment provides a chain-network fusion architecture resource scheduling method for maximizing computing efficiency, comprising the following steps: Step 1: Obtain and save all path structure information of the DAG task. For all DAG tasks to be scheduled in the system, extract all path information (from the source node to the sink node) in each DAG, and save the subtask list contained in each path and its dependency relationship for subsequent critical path analysis. DAG task G1 contains four edges {(A, B), (B, C), (C, D), (A, D)}, which means it contains two subtask dependency paths A->B->C->D and A->D. Task G2 contains four edges {(E, H), (E, F), (F, G), (G, H)}, which means it contains two dependency paths E->H and E->F->G->H.

[0028] Step 2: Calculate the maximum residual computing amount and obtain the critical path of the DAG. For each path in the DAG task, calculate the residual computing amount, which is the sum of the total computing amount of all subtasks on the path that have not been scheduled. The critical path is defined as the path with the maximum residual computing amount. Let p 1,1 represent path A->B->C->D, p 1,2 represent path A->D, p 2,1 represent path E->H, and p 2,2 represent path E->F->G->H, and let the computing amount of each node be 1 10 9FLOPS, and only the sub-tasks represented by E-nodes have been scheduled and executed, then p 1,1 has a remaining computation of 4 10 9 FLOPS, p 1,2 has a remaining computation of 2 10 9 FLOPS, p 2,1 has a remaining computation of 1 10 9 FLOPS, p 2,2 has a remaining computation of 3 10 9 FLOPS. At this time, the path p 1,1， , i.e., A->B->C->D, has the largest remaining computation Li and is identified as the critical path.

[0029] Step 3: Calculate the urgency of all DAGs according to the remaining time. The urgency U of each DAG task is defined as the ratio of the remaining computation on the critical path to the remaining time Tremain. The larger the urgency, the more "urgent" the task is. Taking two DAG tasks G1 and G2 as examples, assuming that the current time is 2 seconds, the deadline of G1 is 10 seconds, and the deadline of G2 is 20 seconds, it can be known that the remaining time of G1 is 8 seconds, and the remaining time of G2 is 18 seconds, so the urgency of p 1,1 is 4 10 9 FLOPS / 8 seconds, the urgency of p 1,2 is 2 10 9 FLOPS / 8 seconds, the urgency of p 2,1 is 1 10 9 FLOPS / 18 seconds, the urgency of p 2,2 is 3 10 9 FLOPS / 18 seconds, among all paths, the urgency of p 1,1 is the highest, which is the critical path.

[0030] Step 4: Evaluate the scheduling of high-urgency sub-tasks on different servers. According to the urgency value calculated in the previous round, the first schedulable (all dependent tasks have been completed) sub-task is selected from the corresponding path. For the selected sub-task, the waiting time, execution time, and expected completion time are evaluated on each candidate server. The result of the previous step is p 1,1The task with the highest urgency is the critical path, therefore, its first node A to be executed is selected for scheduling. Taking servers S1, S2, and S3 as an example, if S1 is idle at the current time, and S2 and S3 need to wait for 1 second, then the waiting time for this task on S1 is considered to be 0 seconds, and the waiting time on S2 and S3 is considered to be 1 second. The computational load of the subtask represented by node A is 10. 9 FLOPS, S1's computing power is 5 10 9 FLOPS / second, S2's computing power is 8 10 9 FLOPS / second, S3's computing power is 4 10 9 If the execution time of the subtask is FLOPS / second, then the execution time of the subtask on S1 is 0.2 seconds, the execution time on S2 is 0.125 seconds, and the execution time on S3 is 0.25 seconds. Since the expected completion time is the sum of the waiting time and the execution time, the expected completion time of the subtask on S1 is 0.2 seconds, the expected completion time on S2 is 1.125 seconds, and the expected completion time on S3 is 1.25 seconds.

[0031] Step 5: Select the server that can complete the task earliest as the scheduling target. Compare the completion times t of the subtasks to be scheduled on each server. finish Select t finish The smallest server is selected as the scheduling target. Based on the calculation results from the previous step, the subtasks on S1, S2, and S3 have different values ​​(t). finish The values ​​of 0.2 seconds, 1.125 seconds, and 1.25 seconds respectively indicate that server S1 can complete the calculation of this subtask the earliest, and therefore the subtask is scheduled to server S1.

[0032] Step 6: Repeat steps 1 to 5 until all subtasks of all DAG tasks have been scheduled. The scheduling process continues, updating the DAG structure, path status, and server resource status in real time until all DAG tasks in the system have been scheduled.

[0033] See details Figure 2 In this embodiment, the application scenarios of the scheduling method are divided into static and dynamic multi-DAG task scheduling. The figure shows a scheduling scenario containing m DAG tasks, described as follows: In the static arrival scenario, all DAG tasks simultaneously enter the ready state at time step 1, allowing direct scheduling of all DAG subtasks. However, in the dynamic arrival scenario, each DAG task has a different arrival time, and at any given moment, only currently arrived subtasks can be scheduled. For example... Figure 2 As shown, in the static case, at any time step t, the DAG set {G1, G2, G3, ..., G...}m All tasks in the static case can be scheduled; while in the dynamic case, due to different arrival times of tasks, only part of the arrived DAGs form the schedulable set before any time step n.

[0034] Embodiment 2 The embodiment provides a chain network fusion architecture resource scheduling system for maximizing computing efficiency, comprising: A path analysis module is configured to acquire full path structure information of all DAG tasks to be scheduled in the system, including a subtask list and a dependency relationship of each path. A critical path calculation module is configured to calculate a residual computing amount of each path, and select a path with a maximum residual computing amount as a critical path. An urgency evaluation module is configured to calculate an urgency value of each DAG task based on the residual computing amount of the critical path and a residual time. A scheduling decision module is configured to select a first schedulable subtask in which all preceding tasks are completed from the critical path with the highest urgency value, and evaluate an expected completion time of the schedulable subtask on each candidate server. A scheduling target determination module is configured to select a server capable of completing the task earliest as a scheduling target. A state updating module is configured to update a task state and a server resource state in real time.

[0035] Further, the execution logic of the scheduling decision module comprises: Based on the urgency value U The schedulable DAG tasks are sorted. The first schedulable subtask on the critical path of the selected DAG is selected. An expected completion time sorting list of the schedulable subtask on each candidate server is generated according to a server waiting time and a computing capability. The subtask is allocated to the server with the minimum expected completion time.

[0036] The above merely provides the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any changes or replacements within the technical range disclosed by the present application can be easily thought of by those skilled in the art, and should be covered within the protection scope of the present application.

[0037] It should be understood that parts not elaborated in the specification are all prior art.

[0038] It should be understood that the above description is merely a detailed explanation of the preferred embodiments and is not intended to limit the patent protection scope of the present application. Any modification or alternation made by those skilled in the art without departing from the scope of the present application shall fall within the patent protection scope of the present application. The patent protection scope of the present application shall be subject to the appended claims.

Claims

1. A chain network fusion architecture resource scheduling method for maximizing computing efficiency, characterized in that, The method comprises the following steps: Step 1: obtaining full path structure information of all DAG tasks to be scheduled in the system, including a subtask list and a dependency relationship of each path; Step 2: calculating a residual computing amount of each path, and selecting a path with a maximum residual computing amount as a critical path; Step 3: calculating an urgency value of each DAG task based on a residual computing amount and a residual time of the critical path; Step 4: selecting a first schedulable subtask with all preceding tasks completed from the critical path with the highest urgency value; and evaluating an expected completion time of the schedulable subtask on each candidate server; Step 5: selecting a server capable of completing the task earliest as a scheduling target; Step 6: repeating steps 1 to 5 until all subtasks of all DAG tasks are scheduled. 2.The resource scheduling method of chain-network fusion architecture for maximizing computing efficiency according to claim 1, wherein, The calculation formula of the residual computing amount of each path in step 2 is: wherein, represents the remaining computation amount of the path P i , represents the computation amount of the subtask in the path , represents the set of subtasks that have not been scheduled. 3.The resource scheduling method of chain-network fusion architecture for maximizing computing efficiency according to claim 1, wherein, The calculation formula of the urgency value in step 3 is: wherein, represents a value of urgency of each DAG task; is the remaining amount of computation for the critical path; is the remaining duration of time for the task from its deadline.

4. The resource scheduling method of chain network fusion architecture for maximizing computing efficiency according to claim 3, characterized in that, A remaining length of time until a deadline of a task The formula for calculating the time is as follows: wherein, denotes the latest completion time instant allowed for the DAG task, denotes the current scheduling time.

5. The resource scheduling method of chain network fusion architecture for maximizing computing efficiency according to claim 1, characterized in that, The selection rule of the schedulable subtask is: On the critical path with the highest urgency value, a first subtask with all dependent tasks completed and not yet executed is selected.

6. The resource scheduling method of chain network fusion architecture for maximizing computing efficiency according to claim 1, characterized in that, The calculation formula of the expected completion time of the schedulable subtask on each candidate server in step 4 is: wherein, is the expected completion time of the schedulable subtask on each candidate server; denotes the waiting time of the schedulable subtask on the candidate server ; denotes the execution time of the schedulable subtask on the candidate server .

7. The resource scheduling method of chain network fusion architecture for maximizing computing efficiency according to claim 6, characterized in that, The step 5 includes: comparing expected completion time of the to-be-scheduled subtask on each server , selecting the smallest server as the scheduling target.

8. The resource scheduling method of chain network fusion architecture for maximizing computing efficiency according to claim 1, characterized in that, Step 6 further comprises: updating the following states in real time after completing the scheduling of each subtask: A set of scheduled subtasks in a DAG path; path remaining computation amount Li and critical path; A waiting queue state of each server.

9. A chain network fusion architecture resource scheduling system oriented to maximize computing efficiency, characterized in that, The method comprises: A path analysis module: used for obtaining full path structure information of all DAG tasks to be scheduled in the system, including a subtask list and a dependency relationship of each path; A critical path calculation module: used for calculating a residual computing amount of each path, and selecting a path with a maximum residual computing amount as a critical path; An urgency evaluation module: used for calculating an urgency value of each DAG task based on a residual computing amount and a residual time of the critical path; A scheduling decision module: used for selecting a first schedulable subtask with all preceding tasks completed from the critical path with the highest urgency value; and evaluating an expected completion time of the schedulable subtask on each candidate server; A scheduling target determination module: used for selecting a server capable of completing the task earliest as a scheduling target; A state updating module: used for updating a task state and a server resource state in real time; The chain-network fusion architecture resource scheduling system for maximizing computing efficiency is used to execute the steps in the chain-network fusion architecture resource scheduling method for maximizing computing efficiency in any one of claims 1 to 8.

10. The chain network fusion architecture resource scheduling system for maximizing computing efficiency according to claim 9, wherein, The execution logic of the scheduling decision module comprises: Based on urgency value U Ranking selects DAG tasks to be scheduled; Selecting a first schedulable subtask on a critical path of a selected DAG; Generating an expected completion time sorting list of the schedulable subtask on each candidate server according to a server waiting time and a computing capability; Assigning the subtask to a server with a minimum expected completion time.

Images