Workflow fault-tolerant scheduling method of end-side cloud cooperative system based on reliability driving

Abstract

The invention discloses a workflow fault-tolerant scheduling method of an end-side cloud cooperative system based on reliability driving. The workflow fault-tolerant scheduling method comprises the following steps: step S1, reading related information of a workflow and resource information of each server; s2, performing scheduling and resource allocation on each task in the workflow by using a fault-tolerant scheduling method to obtain a scheduling scheme; and S3, each server creates a corresponding number of data storage copies, data transmission copies and task execution copies according to the scheduling scheme, and executes the workflow according to the scheduling scheme. The technical problem to be solved by the invention is that in an end-side cloud cooperative computing environment, a workflow can be executed with high reliability through a fault-tolerant mode of creating a data storage copy, a data transmission copy and a task execution copy; the efficient scheduling scheme of the workflow is obtained under the constraint of the dependency relationship and deadline between tasks of the workflow application, the computing resources of the edge server and the total cost of the system, and the reliability of the end-side cloud collaboration system is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of data transmission and data backup in an end-edge-cloud collaborative computing architecture, and in particular to a workflow fault-tolerant scheduling method in an end-edge-cloud collaborative system. Background Art

[0002] With the advent of the Internet of Everything era, the number and types of mobile terminal access devices are growing exponentially, and a large amount of data is collected on terminal devices with heavy computing loads. However, due to the limitation of physical size, the computing resources and battery capacity of mobile terminals are limited. The increased latency will lead to higher energy consumption, and the local computing mode is difficult to meet the low latency requirements of applications. In view of the shortcomings of traditional local processing methods, new computing modes have been proposed, and mobile cloud computing and mobile edge computing technologies have emerged. In the mobile cloud computing (MCC) environment, the computing tasks of terminal devices are offloaded to cloud servers with powerful computing capabilities and processed centrally. Mobile edge computing (MEC) sinks the computing originally concentrated in the cloud to the network edge devices, providing computing services at the end closer to the data source. Compared with edge computing, cloud computing has abundant computing resources, and the computing latency of the cloud is usually smaller than that of edge computing, but the transmission distance is farther than that of the edge server and the transmission latency is higher; while edge computing has the advantages of high bandwidth, low latency and wide range, but it is limited by network edge resources. For the offloading problem under certain complex conditions, it is impossible to solve it by relying on a single computing architecture. In this case, the end-edge-cloud collaborative computing architecture can be an effective solution. By combining the rich computing resources of cloud servers with the low latency of edge servers, the three-tier architecture can provide higher computing and transmission performance than cloud computing or edge computing. However, during the workflow scheduling process, transmission lines, edge servers, and virtual machines are inevitably affected by various failures. Reliability requirements are also an important quality of service (QoS) indicator. In an end-edge-cloud collaborative computing environment, introducing fault-tolerant technology based on active replication into workflow scheduling can effectively enhance the reliability of workflows while meeting the workflow delay constraints, edge server resource constraints, and system total cost constraints.

[0003] In the existing end-edge-cloud collaborative scheduling work, the optimization objectives are mostly latency and energy consumption. Currently, there is no work that considers user reliability in the end-edge-cloud collaborative system. For example, the paper "Collaborative Cloud-Edge-EndTask Offloading in Mobile-Edge Computing Networks With Limited Communication Capability" discloses a task offloading method in an end-edge-cloud collaborative computing environment, whose goal is to minimize the total latency of all mobile devices by determining the computing offloading strategy, computing resources, delivery rate and transmission power allocation under the deadline and energy consumption constraints of the mobile device (MD). The paper "AI-Driven Energy-Efficient Content TaskOffloading in Cloud-Edge-End Cooperation Networks" discloses a task offloading scheme in an end-edge-cloud collaborative network environment based on deep reinforcement learning (DRL), which makes collaborative caching and task offloading decisions in each time slot based on the content request information in the previous time slot and the current network status, thereby minimizing the total energy consumption of the system. The paper "Collaborative cloud-edge-end task offloading with task dependencybased on deep reinforcement learning" discloses a workflow offloading method in the edge-cloud collaborative scenario, which minimizes the average time delay and average energy consumption of all IoT devices under the constraints of IoT devices, multi-core edge servers, and task dependencies. None of the above methods consider fault-tolerant scheduling of workflows under the edge-cloud collaborative architecture.

[0004] The existing task scheduling schemes in the edge-cloud collaborative computing environment do not fully consider the reliability requirements of users for workflow execution, especially the failures that may occur during data storage, data transmission, and task execution. Under the constraints of workflow delay limit, edge server resource limit, and system total cost limit, the reliability of the system needs to be improved in order for the workflow to be successfully executed. Summary of the invention

[0005] In response to the above problems, the present invention provides a reliability-driven workflow fault-tolerant scheduling method for an end-edge-cloud collaborative system, which aims to effectively improve the reliability of workflow execution while ensuring that the partial order relationship between tasks in the workflow, deadlines, cost constraints and computing resource constraints of edge servers are simultaneously met.

[0006] Technical solution:

[0007] A workflow fault-tolerant scheduling method for an edge-cloud collaborative system based on reliability drive is proposed. It obtains the copy allocation strategy by using a fault-tolerant workflow scheduling algorithm that creates data storage copies, data transmission copies, and task execution copies. The data copies and task copies are allocated to different servers and virtual machines (VMs) according to the obtained scheduling scheme, thereby improving the reliability of the entire workflow execution. The main steps of the method include:

[0008] Step S1, read the relevant information of the workflow (including the data volume of each task in the workflow and the average number of computing cycles required for each MB of data volume, the partial order relationship between tasks in the workflow, deadlines, cost constraints, and reliability requirements during data storage, data transmission, and task execution) and the resource information of each server (including the online rate of data storage of each server and the unit price of data storage, the probability of successful data transmission between the edge server and the cloud data center, the bandwidth of data transmission, the unit price of data transmission, the failure rate of each VM on the server, the computing frequency, the unit price of task execution, etc.);

[0009] Step S2, using a fault-tolerant scheduling method to schedule and allocate resources for each task in the workflow, and obtain a scheduling plan (the plan includes a data storage copy, a data transmission copy, a deployment decision plan for a task execution copy, and a minimum total system cost), wherein the fault-tolerant scheduling method includes a task selection method (see step S21 for details) and a copy creation method (see steps S23-S29 for details);

[0010] Step S3: Each server creates a corresponding number of data storage copies, data transmission copies, and task execution copies according to the scheduling plan, and executes the workflow according to the scheduling plan.

[0011] In step S2, each task in the task sequence is scheduled and resource allocated using a fault-tolerant scheduling method to obtain a scheduling solution, which includes the following main steps:

[0012] Step S21, selecting the first task in the to-be-completed workflow task pool as the current task, and parsing the current workflow task information;

[0013] Step S22, judging whether the current task can be executed, the judging standard of whether the current task can be executed is that all the predecessor tasks of the current task have been executed;

[0014] Step S23, initializing the current task and virtual machine pool;

[0015] Step S24, calculate the data transmission time, transmission overhead and transmission reliability required for the current task to deploy data transmission copies on each server, and determine whether the current task is a starting task. If it is a starting task, the starting data transmission copy is the input data generated on the IoT device that is unloaded to the edge server or first unloaded to the edge server and then transmitted to the cloud data center. Otherwise, the data transmission copy of the current task is the data transmission copy of the previous task required to be executed on each server;

[0016] Step S25, calculating the preparation time for deploying task copies of the current task on each server;

[0017] Step S26, calculating the storage overhead and storage reliability of deploying data storage copies of the current task on each server;

[0018] Step S27, calculating the replica deployment budget of the current task;

[0019] Step S28, determine whether the number of task copies of the current task deployment exceeds the maximum number of copies. If not, traverse the virtual machine pool, select the VM deployment task execution copy that meets the task budget constraint and deadline constraint and has the highest reliability, and update the virtual machine pool, task budget and task scheduling reliability, and then go to step S28; if it exceeds the maximum number of copies, the current task scheduling is completed, and go to step S29;

[0020] Step S29, the current task is scheduled and removed from the to-be-completed task pool;

[0021] Step S210, determine whether the task pool to be completed after scheduling is empty, if it is empty, the task scheduling is completed and go to step S211, otherwise go to step S21;

[0022] Step S211, outputting a scheduling plan (replica deployment plan).

[0023] The task selection method in step S21 selects the first task in the to-be-completed workflow task pool as the current task. ij , parse the current workflow task information, including but not limited to the status information of all predecessor tasks of the current task, the data volume L of the task ij , the average number of computing cycles required for each bit of data in the task is C ij wait;

[0024] In the task selection method of step S21, the user may randomly submit multiple workflows to the end-edge-cloud collaborative system. The end-edge-cloud collaborative system includes a set of |m| terminal mobile devices MD={MD1, MD2, ..., MD |m|}, the set of |b| edge base stations BS = {BS1, BS2, ..., BS |b|}, and a remote cloud data center DC d , workflow application set W = {W1, W2, ..., W |W|}, unfinished tasks are put into the to-be-completed workflow task pool. Each workflow can be described as W i ={T i , E i , D i}, respectively representing workflow W i The task set, edge set between tasks and deadline requirements of the i-th application can be used express;

[0025] The executable task judgment method of step S22 determines whether the current task can be executed. It is necessary to traverse all its predecessor tasks to see whether they have been executed. When all the predecessor tasks are completed, the current task can be executed and go to step S23. If the current task has an unfinished predecessor task, the current task enters the end of the queue of the task pool to be completed and goes to step S21.

[0026] In step S23, the reliability of the current task and the cost of the current task are initialized to 0, and the virtual machine pool is initialized so that all virtual machines on the servers can deploy task execution copies;

[0027] Step S24 calculates the data transmission time, transmission overhead and transmission reliability required for the current task to deploy data transmission copies on each server. The present invention studies the workflow application of a single data source. There is only one node in the workflow model as the starting node, that is, the workflow W i Medium i1 The input data of the task is divided into two types: the input data of the starting task is the data generated by the IoT device, and the non-starting task takes the output data of its direct predecessor task as input data;

[0028] Workflow W of step S24 i The starting task t i1 Input data generated on IoT devices i1 The transfer time and its data storage copy Deployment decisions related. Indicates that the data copy is not offloaded through the base station BS b , Indicates that the data copy is offloaded to the BS b and in BS b Deploy data replicas, Indicates that data copies are passed through BS b Offload tasks and transfer data to the cloud data center DC dDeploy data replicas;

[0029] If the data replica is deployed at the base station BS in step S24 b , then the required data transmission time is calculated according to Shannon's law, and the data unloading rate and time Calculated as

[0030]

[0031] Among them B m,b is the channel bandwidth, g0 is the path loss constant, d0 is the reference distance, d is the actual distance from the user terminal device to the edge server, θ is the path loss index, p is the transmission power of the terminal device, and N0 is the noise power spectral density between the terminal device and the edge server;

[0032] Data transmission overhead and data transmission reliability Respectively expressed as

[0033]

[0034] in, and Respectively represent MD m to BS b The transmission unit price and the probability of successful transmission.

[0035] If the data replica is deployed in the base station DC in step S24 d , its data transmission also includes from BS b To DC d The data transmission time, overhead and reliability are calculated as follows:

[0036]

[0037] Among them, BW b,d , and Respectively represent from BS b To DC d bandwidth, transmission unit price and probability of successful transmission;

[0038] Workflow W of step S24 i Non-starting task t ij The data is stored as the output data of its immediate predecessor task, where the predecessor task Output data The transmission and task t ij Data storage replication decisions The data transmission time, overhead and probability of successful transmission are expressed as:

[0039]

[0040] If any copy of the data is successfully transmitted, the transmission of this data is reliable. The transmission reliability of is calculated as:

[0041]

[0042] Where |dt| represents The total number of data store replicas deployed;

[0043] Data ij On the server s The total transmission reliability of the deployed data replica is calculated by multiplying the transmission reliability of all predecessor tasks:

[0044]

[0045] Computational data storage replicas are deployed on the server s Transmission cost Including the overhead of deploying the transmission replicas of all predecessor tasks, the specific calculation is:

[0046]

[0047] The calculation of the preparation cost and preparation time of deploying the task replicas of the current task on each server in step S25 is related to the deployment of the data transmission replicas;

[0048] The starting task t of step S25 i1 On the server s Preparation time for execution Calculated as:

[0049]

[0050] The non-starting task task t of step S25 ij On the server s Preparation time for execution Completion time of the predecessor task copy Related to the transfer time of the transferred copy:

[0051]

[0052] The calculation of the storage overhead of deploying data storage copies on each server in step S26 is related to the storage reliability and deployment of data storage copies. The data storage overhead is calculated as

[0053]

[0054] in For server s The storage unit price;

[0055] The reliability of data storage in step S26 is defined as the probability of successful data storage. Data storage reliability is related to the server selected for data replica deployment:

[0056] d ij The data copy is deployed on the edge server BS b When , the reliability of data storage is related to the reliability of the edge server. The edge server changes its state with the working time, and its reliability follows an exponential distribution that decays with the increase of working time. b Indicates BS b The online rate and data storage reliability are calculated as:

[0057]

[0058] d ij The data copy is deployed on the cloud server DC d The reliability of data storage is related to the reliability of the cloud server. d Online rate represents the probability of its accessibility, then the reliability of data storage is expressed as:

[0059]

[0060] Step S27 calculates the replica deployment budget of the current task. To ensure that the starting task can be successfully offloaded to the base station, the budget of the workflow application i Divided into minimum offload data budget and disposable budget Right now The task scheduling budget for the current task is calculated as:

[0061]

[0062] Among them, Act (Budget i ) is the workflow W i The actual remaining budget, NS(L i ) is W i The total amount of data for tasks that have not yet been assigned;

[0063] Task execution copy of step S28 On the server s The kth virtual machine vm s,k Execution time and execution overhead Calculated as

[0064]

[0065] Among them, F s,k and vm s,k The calculation frequency and task execution unit price;

[0066] The calculation of the task execution replica reliability in step S28 is defined as the probability of successful task execution. The present invention takes into account the temporary failure of the VM. ij The probability of a transient failure during execution follows a Poisson distribution, denoted by λs ,k Represents vm s,k The failure rate of task t ij In vm s,k The reliability of the deployment task execution replica is calculated as:

[0067]

[0068] Step S28 updates the state of the VM of the currently deployed execution copy, including updating the idle time slot of the VM, and removing the VM of the currently deployed task execution copy from the virtual machine pool where the task copy can be deployed;

[0069] The updating current task scheduling overhead of step S28 includes the overhead of deploying all data transmission copies, data storage copies, and task execution copies;

[0070] Step S28 ij The reliability of task scheduling is related to the reliability of data transmission replicas, data storage replicas, and task execution replicas. The task replica can only be successfully executed if the data replica is successfully transmitted and stored. Therefore, the task replica The probability of successful scheduling can be expressed as:

[0071]

[0072] The probability of task replica scheduling failure is:

[0073]

[0074] The present invention uses an active replication scheme to perform fault tolerance on task execution replicas. ij If any of the copies of the task is successfully completed, then the execution of the task has no failures. The task t with |te| task scheduling copies ij Reliability is calculated as

[0075]

[0076] The output scheduling plan (replica deployment plan) of step S211 includes the deployment decision plan of the data storage copy, data transmission copy, task execution copy and the reliability of the end-edge-cloud collaborative system;

[0077] The reliability of the end-edge-cloud collaborative system in step S211 is the sum of the reliabilities of all workflow scheduling in the system, and the reliability of a single workflow scheduling is the product of the reliabilities of all tasks in the workflow.

[0078] Beneficial Effects of the Invention

[0079] The technical problem solved by the present invention is to enable workflows to be executed with high reliability in an end-edge-cloud collaborative computing environment by creating fault-tolerant copies of data storage, data transmission and task execution. An efficient scheduling solution for workflows is obtained under the constraints of the dependencies and deadlines between tasks in the workflow application, the computing resources of the edge server and the total cost of the system, thereby effectively improving the reliability of the end-edge-cloud collaborative system. BRIEF DESCRIPTION OF THE DRAWINGS

[0080] Figure 1 It is a schematic diagram of the overall process of a workflow fault-tolerant scheduling method of an end-edge-cloud collaborative system in a specific embodiment of the present invention.

[0081] Figure 2 It is a specific flow chart of the fault-tolerant scheduling method in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0082] Combine the following Figure 1 The embodiment is elaborated in detail. In this embodiment, there are two users uploading workflow applications. Each user workflow has three tasks to be processed. These six tasks are required to complete workflow scheduling in an end-edge-cloud collaborative system with two edge servers, one cloud data center, and nine VMs and obtain a scheduling plan with the minimum total system cost.

[0083] The basic information of the task is as follows:

[0084]

[0085] The cost constraint of workflow application 0 is $10.33, and the cost constraint of workflow application 1 is $17.65.

[0086] The basic information of the server is as follows:

[0087]

[0088]

[0089] The transmission price from the terminal device to the edge server is 0.035$ / GB / s, and the transmission price from the edge server to the cloud data center is 0.5$ / GB / s. The transmission reliability information between servers is as follows:

[0090] Transmission reliability Server 0 Server 1 Server 2 Server 0 0.99 0.98 0.95 Server 1 0.98 0.99 0.91 Server 2 0.95 0.91 0.99

[0091] The basic information of the VM is as follows:

[0092]

[0093] Combination Figure 2 As shown in the flowchart, a workflow fault-tolerant scheduling method for a reliability-driven end-edge-cloud collaborative system specifically includes the following steps:

[0094] Step S1, read the relevant information of the workflow, including the data volume of each task in the workflow and the average number of computing cycles required for each MB of data volume, the partial order relationship between tasks in the workflow, etc., read the resource information of each server, including the online rate and unit price of data storage of the server data, the probability of successful data transmission between servers, the bandwidth of data transmission, the unit price of data transmission, the failure rate of each VM on the server, the computing frequency, the unit price of task execution, etc., define the task set as π = {π0, π1, π2, π3, π4, π5}, and establish the relevant scheduling model;

[0095] Step S2, using a fault-tolerant scheduling method to schedule and allocate resources for each task in the workflow, and obtain a scheduling scheme (the scheme includes a deployment decision scheme for a data transmission copy, a data storage copy, a task execution copy, and a maximum system reliability), wherein the fault-tolerant scheduling method includes features such as a task selection method and a copy creation method;

[0096] The reliability-driven fault-tolerant scheduling method in step S2 schedules each task in the workflow and allocates resources to obtain a scheduling solution, which includes the following main steps:

[0097] The task selection method in step S21 selects the first task in the to-be-completed workflow task pool as the current task and parses the current workflow task information;

[0098] The executable task determination method in step S22 is based on the determination criteria that all predecessor tasks of the current task have been completed;

[0099] Step S23, initializing the current task and virtual machine pool;

[0100] Step S24, calculate the data transmission time, transmission overhead and transmission reliability required for the current task to deploy data transmission copies on each server, and determine whether the current task is a starting task. If it is a starting task, the starting data transmission copy is the input data generated on the IoT device that is unloaded to the edge server or first unloaded to the edge server and then transmitted to the cloud data center. The data transmission time, overhead and reliability are calculated according to formula 2-7. Otherwise, the data transmission copy of the current task is the data transmission copy of the predecessor task required to be executed on each server. The data transmission reliability and overhead are calculated according to formula 12 and formula 13 respectively.

[0101] Step S25, calculating the preparation time for deploying task copies of the current task on each server according to Formula 14 and Formula 15;

[0102] Step S26, calculating the storage overhead and storage reliability of deploying data storage copies on each server for the current task according to formulas 16-18;

[0103] Step S27, calculating the replica deployment budget of the current task according to Formula 19;

[0104] Step S28, determine whether the number of task copies deployed by the current task exceeds the maximum number of copies. If not, traverse the virtual machine pool, select the VM deployment task execution copy that meets the task budget constraint and deadline constraint and has the highest reliability, and update the virtual machine pool and task budget. Calculate the scheduling reliability of the task according to formula 25, and then go to step S28; if it exceeds the maximum number of copies, the current task scheduling is completed, and go to step S29;

[0105] Step S29, the current task is scheduled and removed from the to-be-completed task pool;

[0106] Step S210, determine whether the task pool to be completed after scheduling is empty, if it is empty, the task scheduling is completed and go to step S211, otherwise go to step S21;

[0107] Step S211, outputting a scheduling plan (a replica deployment plan and a maximum system reliability of 1.9706);

[0108] Step S3: Each server creates a corresponding number of data storage copies, data transmission copies, and task execution copies according to the scheduling plan, and executes the workflow according to the scheduling plan.

[0109] This application proposal introduces a workflow fault-tolerant scheduling method for an end-edge-cloud collaborative system based on reliability-driven, ensuring that the system reliability is maximized under the conditions of satisfying the partial order relationship between tasks in the workflow, deadlines, cost constraints, and computing resource constraints of edge servers. Existing task scheduling schemes in end-edge-cloud collaborative computing environments do not fully consider users' reliability requirements for workflow execution, especially failures that may occur during data storage, data transmission, and task execution, which will cause the execution of the workflow to fail to complete smoothly. The method described in this application proposal comprehensively considers the reliability requirements of workflows in data storage, data transmission, and task execution. It not only uses data storage replicas and task execution replicas, but also creates data transmission replicas through active replication. The system solves the problem of workflow fault-tolerant scheduling in end-edge-cloud collaborative environments.

[0110] The present invention provides a workflow fault-tolerant scheduling method for a reliability-driven end-edge-cloud collaborative system. There are many methods and ways to implement the technical solution. The above is only a preferred implementation of the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principle of the present invention, several improvements and modifications can be made, and these improvements and modifications should also be regarded as the scope of protection of the present invention. All components not specified in this embodiment can be implemented using existing technologies.

Claims

1. A workflow fault-tolerant scheduling method for a reliability-driven end-edge-cloud collaborative system, characterized in that The following steps are involved: Step S1, read the relevant information of the workflow and the resource information of each server; the relevant information of the workflow includes but is not limited to the data volume of each task in the workflow and the average number of computing cycles required for each MB of data volume, the partial order relationship between tasks in the workflow, deadline constraints, and reliability requirements during data storage, data transmission, and task execution; the server resource information includes but is not limited to the online rate of data storage of each server and the unit price of data storage, the probability of successful data transmission between the edge server and the cloud data center, the bandwidth of data transmission, the unit price of data transmission, the failure rate of each VM on the server, the computing frequency, and the unit price of task execution; Step S2, using a fault-tolerant scheduling method to schedule and allocate resources for each task in the workflow, and obtain a scheduling plan, which includes a deployment decision plan for a data storage copy, a data transmission copy, a task execution copy, and a minimum total system cost of the task; the fault-tolerant scheduling method includes a task selection method and a copy creation method; Step S3: Each server creates a corresponding number of data storage copies, data transmission copies, and task execution copies according to the scheduling plan, and executes the workflow according to the scheduling plan.

2. The method according to claim 1, characterized in that The fault-tolerant scheduling algorithm includes the following steps: Step S21, a task selection method, selecting the first task in the to-be-completed workflow task pool as the current task, and parsing the current workflow task information; Step S22, an executable task determination method is used to determine whether the current task is executable; Step S23, initializing the current task and virtual machine pool; Step S24, calculate the data transmission time, transmission overhead and transmission reliability required for the current task to deploy data transmission copies on each server, and determine whether the current task is a starting task. If it is a starting task, the starting data transmission copy is the input data generated on the IoT device that is unloaded to the edge server or first unloaded to the edge server and then transmitted to the cloud data center. Otherwise, the data transmission copy of the current task is the data transmission copy of the previous task required to be executed on each server; Step S25, calculating the preparation time for deploying task copies of the current task on each server; Step S26, calculating the storage overhead and storage reliability of deploying data storage copies of the current task on each server; Step S27, calculating the replica deployment budget of the current task; Step S28, determine whether the number of task copies of the current task deployment exceeds the maximum number of copies. If not, traverse the virtual machine pool, select the VM deployment task execution copy that meets the task budget constraint and deadline constraint and has the highest reliability, and update the virtual machine pool, task budget and task scheduling reliability, and then go to step S28; if it exceeds the maximum number of copies, the current task scheduling is completed, and go to step S29; Step S29, the current task is scheduled and removed from the to-be-completed task pool; Step S210, determine whether the task pool to be completed after scheduling is empty, if it is empty, the task scheduling is completed and go to step S211, otherwise go to step S21; Step S211, output the scheduling plan.

3. The method according to claim 2, characterized in that The parsing information in step S21 includes but is not limited to parsing the current workflow task information and all virtual machine information in the virtual machine pool. The workflow task information includes but is not limited to the status information of all previous tasks of the current task, the data volume L of the task ij , the number of CPU cycles required for each MB of data in the task is C ij .

4. The method according to claim 2, characterized in that: In step S22, to determine whether the current task can be executed, it is necessary to traverse all its predecessor tasks to see if they have completed preparations. When all the predecessor tasks are completed, the current task can be executed and go to step S23. If the current task has an unfinished predecessor task, the current task enters the end of the queue of the to-be-completed task pool and goes to step S21.

5. The method according to claim 2, characterized in that: In step S23, the reliability of the current task and the cost of the current task are initialized to 0, and the virtual machine pool is initialized so that the virtual machines of all servers can deploy task execution copies.

6. The method according to claim 2, characterized in that In step S24: There are two types of task input data: the input data of the workflow's starting task is the data generated by the IoT device, and the non-starting task uses the output data of its direct predecessor task as input data; ① If the data storage replica is deployed at the edge base station, then data transmission only needs to offload the task to the edge base station. At this time: Data offloading rate to edge base stations Calculated according to the following formula: Among them, B m,b is the channel bandwidth, g0 is the path loss constant, d0 is the reference distance, d is the actual distance from the user terminal device to the edge server, θ is the path loss index, p is the transmission power of the terminal device, and N0 is the noise power spectral density between the terminal device and the edge server; Calculated according to the following formula: in, Represents a data store replica By mobile device MD m Data transmission time offloaded to the edge base station BS, L ij Indicates the amount of data for the current task; and Obtained by the following formula: in, and Respectively represent MD m to BS b The transmission unit price and the probability of successful transmission; Represents a data store replica By mobile device MD m Offload to edge base station BS b Data transmission overhead; Represents a data store replica By mobile device MD m Offload to edge base station BS b Data transmission reliability; ② If the data storage replica is deployed in the cloud data center, the data transmission needs to be offloaded to the edge base station first, and then transmitted to the cloud data center through the edge base station. At this time: Among them, BW b,d , and Respectively represent from BS b To DC d bandwidth, transmission unit price and probability of successful transmission; Represents a data store replica By mobile device MD m Offload to edge base station BS b Then to the cloud data center DC d Data transmission time; Represents a data store replica By mobile device MD m Offload to edge base station BS b Then to the cloud data center DC d Data transmission overhead; Represents a data store replica By mobile device MD m Offload to edge base station BS b Then to the cloud data center DC d Data transmission reliability; ③Non-starting task t ij The data storage of is the output data of all its direct predecessor tasks. At this time, the transmission of the output data of a single predecessor task is related to the data storage copy decision of the current task. If any data copy of the data transmits the data successfully, then the transmission of this data is reliable. At this time: in, Represents a data transfer copy By server Transfer to server s Data transmission time, represents the transmission overhead of data transmission replicas, Indicates the transmission reliability of the data transmission copy, and They represent data transmission bandwidth, transmission unit price and probability of successful transmission respectively; data The transmission reliability of is obtained by the following formula: Where |dt| represents The total number of data store replicas deployed; The total transmission reliability of the data replica deployed by the non-starting task is the product of the transmission reliability of all the predecessor tasks, and the total transmission overhead The cost of transferring replica deployments for all predecessor tasks is obtained as follows:

7. The method according to claim 2, characterized in that: In step S25: ① The starting task of the workflow i1 Preparation time for deploying data storage replicas on the server The time for transmitting input data on the server is obtained by the following formula: in, Indicates that the data copy is offloaded to the BS b and in BS b Deploy data replicas, Indicates that data copies are passed through BS b Offload tasks and transfer data to the cloud data center DC d Deploy data replicas; ② Non-starting task t ij Preparation time for deploying data storage replicas on the server Completion time of the predecessor task copy It is related to the transmission time of the transmission copy and is obtained by the following formula: in, Indicates the latest completion time of the task execution copies of all predecessor tasks.

8. The method according to claim 2, characterized in that: In step S26: ① Storage overhead is the product of storage unit price and data volume; ②The reliability of data storage is obtained by the following formula: in, Indicates d ij The data copy is deployed on the edge server BS b Reliability, λ b Indicates BS b Online rate; ③The data storage reliability of the cloud data center is obtained by the following formula: in, Indicates d ij The data copy is deployed on the cloud server DC d Reliability, Indicates cloud data center DC d Online rate.

9. The method according to claim 2, characterized in that: In step S27, the task scheduling budget of the current task is obtained by the following formula: in, To minimize the offload data budget, Act (Budget) is the disposable budget. i ) is the workflow W i The actual remaining budget, NS(L i ) is W i The total amount of data in the tasks that have not yet been assigned.

10. The method according to claim 2, characterized in that In step S28: ①Task execution copy On the server s The kth virtual machine vm s,k Execution time and execution overhead Calculated as: Among them, F s,k and vm s,k The calculation frequency and task execution unit price; ②Task execution reliability is defined as the probability of successful task execution. ij The probability of a transient failure during execution follows a Poisson distribution, and the task t ij In vm s,k Reliability of deployment task execution replicas Calculated by the following formula: Among them, λ s,k Represents vm s,k failure rate; ③Task t with |te| task scheduling copies ij Reliability ts (t ij ) is represented by the following formula: in, Represents a task copy The probability of successful scheduling is obtained by the following formula: in, and Respectively represent the server s Deploy data store replicas data transmission reliability and data storage reliability, Represents a task copy In vm s,k Reliability of execution.

Images