Task allocation and scheduling method based on multi-round auction

Abstract

The invention relates to the technical field of communication, in particular to a task allocation and scheduling method based on multiple rounds of auction, which constructs an auction framework for a scene in which terminal equipment simultaneously processes multiple tasks in parallel and based on a D2D collaborative MEC architecture, and takes a busy terminal as a buyer, an idle terminal as a seller and a trusted third party as an auctioneer. Edge collaborative computing unloading based on an incentive mechanism and task scheduling are combined to improve system utility, specifically, a resource provider selection strategy based on a bidding performance ratio and a task scheduling sequence strategy based on a greedy profit-emergency ratio are adopted, and optimal matching and an optimal task processing sequence are achieved. Simulation results show that the number of effective processing tasks and time delay of task processing are increased, the advantages of edge collaboration task unloading and flexible task processing sequence adjustment are fully utilized, and the resource utilization rate is increased.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to a task allocation and scheduling method based on multi-round auctions. Background Art

[0002] With the explosive growth of smart terminal devices and new applications, traditional cloud computing architecture cannot meet the latency requirements of these new applications due to long propagation delays. To solve these problems, Multi-access Edge Computing (MEC) came into being. MEC deploys limited computing resources at the edge of the wireless access network and helps terminal devices process tasks through offloading technology, aiming to achieve low-latency task requests. The latest trend of Multi-access Edge Computing (MEC) is collaborative offloading, that is, the computing tasks of busy terminals (such as smartphones, computers, etc.) can not only be executed by MEC servers, but also offloaded to other available terminal devices nearby (called idle terminals) through technologies such as D2D communication. However, there are some important challenges in realizing collaborative offloading in MEC networks. Executing computing tasks requires consuming computing resources, and idle terminals are selfish and unwilling to contribute their limited computing resources.

[0003] In order to attract idle terminals and edge servers to participate in collaborative computing, it is crucial to introduce an incentive mechanism in the MEC system. However, the existing resource allocation and scheduling methods have introduced incentive mechanisms, but have not considered the collaborative offloading between different terminal devices and the rational use of the available resources of the terminal devices; nor have they considered the impact of the task scheduling order on the task completion delay. In reality, they are only limited to the sorting problem of processing only a single task at the same time, and have not explored the scenario where idle terminals process multiple tasks in parallel at the same time, which wastes computing resources and fails to achieve the goal of maximizing system utility. Summary of the invention

[0004] The purpose of the present invention is to provide a task allocation and scheduling method based on multi-round auctions, which aims at the scenario where terminal devices process multiple tasks simultaneously and in parallel, minimizes delay, energy consumption and penalty costs, increases the number of tasks that are effectively processed, and improves resource utilization.

[0005] To achieve the above object, the present invention provides a task allocation and scheduling method based on multi-round auctions, comprising the following steps:

[0006] Step 1: Build an auction framework including auctioneers, buyers and sellers;

[0007] Step 2: After receiving the task information published by the auctioneer, the resource provider submits a task bid to the auctioneer;

[0008] Step 3: The auctioneer selects the winner of each task based on the bid-performance ratio of each task;

[0009] Step 4: Task scheduling sequence plan based on urgency-profit ratio;

[0010] Step 5: Send the optimal result obtained in step 4 to the task executor;

[0011] Step 6: After the task is completed, the resource provider charges the corresponding buyer and returns its calculation results.

[0012] Optionally, the auctioneer in step 1 is a trusted third party, i.e., a central cloud, an edge cloud, or a professional auction organization, the buyer is a busy terminal, and the seller is an idle terminal. During the auction, all participants are honest and their bids satisfy p m,j (k) = c m,j .

[0013] Optionally, during the execution of step 2, the edge server sends the information of each task (d m ,C m ,t m ,τ m ) is sent to resource providers for auction, where d m is the amount of data for the computing task m∈M, C m is the computing resources required to process the task, τ m is the maximum allowable delay of task m∈M, t m is the time when task m∈M is generated on a busy terminal, M={1,2,...,M} represents the set of busy terminals, C={1,2,...,j,...,C} represents the set of all idle terminals;

[0014] After receiving the task information, the resource provider submits the task bid to the auctioneer. They represent the bid price p of resource provider j∈J for task m∈M in the kth auction round. m,j (k), the computing power provided f j , and the waiting delay required for the task J = {C∪0} represents the set of resource providers for all tasks in the system (index 0 represents the edge server).

[0015] Optionally, the bid performance ratio expression in step 3 is as follows:

[0016]

[0017] Among them, δ∈[0,1] is the control coefficient.

[0018] Optionally, in step 4, each computing task m∈M is first matched with the resource provider with the lowest bid-performance ratio. There are two situations:

[0019] The number of tasks matched by resource provider j∈J is less than or equal to the maximum number of tasks it can process simultaneously;

[0020] The number of tasks matched by resource provider j∈J is greater than the maximum number of tasks n that it can process simultaneously j .

[0021] Optionally, when the number of matching tasks is less than or equal to the maximum number of tasks that the resource provider can process simultaneously, the utility of the system completing the tasks is calculated using the following formula:

[0022]

[0023] Among them, v m is the true estimate of the task, p e,j , p t,j They represent the economic cost of unit energy consumption and delay respectively;

[0024] For U m,j The task with U<0 will be discarded. m,j Tasks ≥ 0 are added to the matching task set M assigned Then update the waiting delay and bidding price on resource provider j∈J for the next auction round. Then the current auction (the kth auction) Will be updated to look like this:

[0025]

[0026] in, Represents the set of tasks assigned in the kth round of auction, and then according to Update bid priceb m,j (k+1), then the corresponding bidding performance ratio γ m,j The values ​​will also be updated to reflect changes in wait delays and bid prices.

[0027] Optionally, when the number of matching tasks is greater than the maximum number of tasks that the resource provider can handle simultaneously, the tasks are sorted according to the profit-urgent ratio λ. m,j Arranged from large to small, resource provider j∈J will select the first n j tasks, if in the first n j tasks have a profit-urgent ratio λ m,j Tasks with a value less than 0 will be discarded directly;

[0028] Profit-Emergency Ratio λ m,jis the ratio of system utility to time urgency, expressed as follows:

[0029]

[0030] The present invention provides a task allocation and scheduling method based on multi-round auctions. Aiming at the scenario where terminal devices process multiple tasks simultaneously and in parallel, a multi-task scheduling sequence model is constructed based on the D2D collaborative MEC architecture. Busy terminals are regarded as buyers, idle terminals are sellers, and third-party central clouds, edge clouds or professional auction institutions are used as auctioneers. The edge collaborative computing offloading and task scheduling based on the incentive mechanism jointly improve the system utility. Specifically, a resource provider selection strategy based on the bidding performance ratio and a task scheduling sequence strategy based on the greedy profit-urgent ratio are adopted to achieve the matching of the best and the best task processing sequence. The simulation results show that the present invention increases the number of effectively processed tasks and the system utility, fully mobilizes and utilizes the advantages of edge collaborative task offloading and flexible adjustment of task processing sequence, and improves resource utilization. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0032] Figure 1 It is a schematic flow chart of the steps of a task allocation and scheduling method based on multi-round auctions of the present invention.

[0033] Figure 2 It is a schematic diagram of the structure of a multi-task scheduling sequence model of a specific embodiment of the present invention.

[0034] Figure 3 It is a schematic diagram of task scheduling in a specific embodiment of the present invention.

[0035] Figure 4 It is a schematic diagram of the penalty function relationship based on delay in a specific embodiment of the present invention.

[0036] Figure 5 It is a schematic diagram of execution time vs. number of tasks in a specific embodiment of the present invention.

[0037] Figure 6 It is a schematic diagram of system utility vs. number of tasks in a specific embodiment of the present invention.

[0038] Figure 7 It is a schematic diagram of the number of effectively processed tasks vs. the number of tasks in a specific embodiment of the present invention.

[0039] Figure 8 It is a schematic diagram comparing the system utility of each method vs. the number of tasks in a specific embodiment of the present invention.

[0040] Fig. 9 It is a schematic diagram comparing the number of tasks effectively processed by each method vs the number of tasks in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0041] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0042] See also Figure 1 The present invention provides a task allocation and scheduling method based on multi-round auctions, comprising the following steps:

[0043] S1: Construct an auction framework including auctioneers, buyers and sellers;

[0044] S2: After receiving the task information published by the auctioneer, the resource provider submits a task bid to the auctioneer;

[0045] S3: The auctioneer selects the winner of each task based on the bid-performance ratio of each task;

[0046] S4: Task scheduling sequence scheme based on urgency-profit ratio;

[0047] S5: Send the optimal result obtained in step S4 to the task executor;

[0048] S6: After the task is completed, the resource provider charges the corresponding buyer and returns its calculation results.

[0049] The following is further described in conjunction with specific embodiments and execution steps:

[0050] like Figure 2As shown, a task allocation and scheduling method based on multi-round auctions of the present invention is specifically executed on an MEC system that can realize D2D offloading and can flexibly adjust the task processing order. The corresponding system is composed of an edge cloud pool (edge ​​server) and multiple terminal devices. The terminal devices include two types: busy terminals that generate task requests and idle terminals with available resources (such as laptops, tablets and other smart devices). Let M = {1, 2, ..., M} represent the busy terminal set, and C = {1, 2, ..., j, ..., C} represent the set of all idle terminals. The computing tasks generated by the busy terminals are offloaded to the edge servers through cellular links or to the idle terminals through D2D links for processing to meet their own task requirements, and then the edge servers and idle terminals sort and process the tasks offloaded thereto in parallel. Let J = {C∪0} represent the set of resource providers for all tasks in the system (index 0 represents the edge server), then the computing power available set of resource providers (f j ,n j ),j∈{0,1,2,...,|J|} represents, f j represents the CPU frequency (clock cycles / second) of resource provider j∈J (a general term for idle terminals and edge servers), n j Indicates the maximum number of tasks that can be processed simultaneously.

[0051] Accordingly, during the execution of the present invention, the system needs to go through the sequential processes of task unloading, task transmission and calculation, and task scheduling.

[0052] The present invention assumes that busy terminals do not process tasks locally due to their limited computing power. To simplify the calculation, each terminal has a computing task to be offloaded, and the tasks are independent and indivisible. The set of tasks to be processed in the system can be expressed as M = {1, 2, ..., M}. The computing task m∈M generated by the busy terminal m∈M is represented by (d m ,C m ,τ m ,t m ) indicates that d m is the data size of the computing task m∈M (in bits), C m is the computing resources required to process the task (in CPU cycles), τ m is the maximum allowable delay of task m∈M, t m is the time when task m∈M is generated on the busy terminal. Introducing the unloading decision variable x m,j (j∈J), where x m,j =1 indicates that task m∈M is offloaded to resource provider j∈J for processing, otherwise x m,j= 0. At this time, the following constraints exist, indicating that the busy terminal can only be unloaded to one resource provider at most:

[0053]

[0054] The busy terminal offloads the generated computing tasks to the edge server or idle terminal for processing. The whole offloading process includes three parts. First, the task is uploaded to the resource provider (edge ​​server or idle terminal) through the wireless channel to wait for processing. Then the resource provider determines the time to start processing the task and allocates computing resources for it to process it. Finally, the result is returned to the user. Since the amount of data returned is much smaller than the amount of uploaded data, the delay and energy consumption in the result return process are not considered in this invention. In summary, the delay of the computing task m∈M Including upload delay Waiting delay And processing delay System energy consumption includes the transmission energy consumption of computing tasks And processing energy consumption

[0055] for The upload delay from offloading to the resource provider can be expressed as:

[0056]

[0057] in, Indicates the uplink transmission rate, w m,j P represents the transmission bandwidth allocated by resource provider j∈J for computing tasks. Each busy terminal uses the same bandwidth for data transmission and only transmits one task at a time. m,j Indicates the uplink transmission power, H m,j is the channel gain, σ 2 represents the channel noise power. At this time, according to the transmission delay, the corresponding transmission energy consumption is:

[0058]

[0059] When task m∈M is uploaded to resource provider j∈J, the resource provider allocates corresponding computing resources to process it. At this time, the processing delay of the task is:

[0060]

[0061] The corresponding computing and processing energy consumption can be obtained, expressed as:

[0062]

[0063] Among them, κ jis the energy consumption coefficient of the idle terminal j∈J, which depends on the effective switching capacitance of the chip architecture.

[0064] Since the resource provider’s own computing resources are limited, it is often unable to process all the tasks that have arrived at the same time. The tasks that arrive later need to wait in the task scheduling buffer (used to cache and schedule tasks that cannot be processed in time). For any resource provider j∈J, we introduce the task scheduling decision variable To represent the scheduling strategy of the task m∈M offloaded to it. Among them, represents the start time of task m∈M, Indicates the time when task processing ends, and there is the following relationship:

[0065]

[0066] For The task to be processed at the next moment will only be executed when all the tasks to be processed at this moment are completed. j tasks (maximum number of processing tasks), so the following constraints exist:

[0067]

[0068] in, Representing a collection Remove tasks For tasks other than , ∩(·) is an indicator function, which takes the value 1 if . is true, otherwise it takes the value 0.

[0069] Next, we will give an example to illustrate the specific process of task sorting. For example, there are five computing tasks in the system that need to be offloaded to resource provider j∈J for processing, and resource provider j∈J can process at most 2 tasks at the same time. The task scheduling strategy is known. The specific scheduling diagram of the task is as follows Figure 3 Example.

[0070] In order to maximize the system utility, the present invention also considers the system utility from two aspects: busy terminals and resource providers, and at the same time models the problem with constraints.

[0071] Busy terminals offload tasks to resource providers (edge ​​servers and idle terminals) for computing, in order to expand computing power and meet task latency requirements, while giving them corresponding rewards. Tasks that fail due to timeouts will be punished. Next, the specific definitions of user utility, resource provider utility, and system utility are given.

[0072] (1) Busy Terminal Utility

[0073] For a busy terminal m∈M, its utility is defined as the value of the task v m Subtract the price paid to resource provider j∈J to get p m,j and the transmission energy consumption required for offloading tasks It can be expressed as:

[0074]

[0075] (2) Utility of resource providers

[0076] Considering that the computing resources available to resource providers are limited and cannot provide satisfactory services for all computing tasks (the present invention uses task delay as an indicator to measure the quality of service QoS), in order to motivate resource providers to meet the delay requirements of tasks to the greatest extent, the present invention designs a penalty function based on delay, that is, when the task processing time exceeds its maximum delay, it will cause a loss in task valuation, which will be converted into a penalty for the resource provider. The penalty function is expressed as follows:

[0077]

[0078] in, The function represents the change of task value with delay, and the expression is as follows:

[0079]

[0080] Among them, τ m is the maximum allowable delay of the task, v m is the true estimate of the task, α is the control delay exceeding τ m The parameter of the curve decline degree. Figure 4 Given a penalty function With delay relationship.

[0081] In summary, the cost of resource provider j∈J processing computing task m∈M can be expressed as:

[0082]

[0083] Among them, m∈M, j∈J and p e,j , p t,j They represent the economic cost of unit energy consumption and delay respectively.

[0084] Then the utility of resource provider j∈J processing computing task m∈M can be expressed as:

[0085]

[0086] (3) System effectiveness

[0087] The present invention defines system utility as the weighted sum of busy terminal utility and resource provider utility, which can be expressed as:

[0088]

[0089] Based on the above description, the present invention maximizes the system utility under the D2D collaborative MEC system architecture while providing users with satisfactory service quality. Therefore, the overall optimization problem can be formulated as the following P1 problem:

[0090]

[0091] P1 contains two types of decision variables, namely, the offloading decision variable X and the task scheduling decision variable θ={t s ,t e Among them, C1 and C2 are offloading decision constraints, indicating that the task can only be offloaded to one resource provider, constraint C3 indicates that the task can only be processed after being uploaded to the resource provider, constraint C4 indicates the processing delay of the task on the resource provider, and constraint C5 indicates that the number of tasks processed in parallel cannot exceed the maximum number of tasks that the resource provider can process simultaneously.

[0092] In problem P1, task allocation and scheduling with the goal of maximizing system utility mainly focuses on two sub-problems: 1) deciding where to offload tasks and 2) determining the processing order on task resource providers.

[0093] Steps S2 to S6 are the detailed process of problem solving. In the multi-round auction process, busy terminals are buyers, idle terminals are sellers, and a trusted third party (central cloud, edge cloud or professional auction organization) is the auctioneer. The edge server sends the information of each task in the system (d m ,C m ,t m ,τ m ) is sent to the resource provider for auction. After receiving the task information, the resource provider submits the task bid to the auctioneer. They represent the bid price p of resource provider j∈J for task m∈M in the kth auction round. m,j (k), the computing power provided f j , and the waiting delay required for the task The present invention assumes that all participants are honest and that their bids satisfy p m,j (k) = c m,j .

[0094] Step S2: Submit Bid

[0095] After receiving the task information, the resource provider submits the task bid to the auctioneer.

[0096] Step S3: Resource provider selection strategy based on bid performance ratio (BPR)

[0097] After receiving bids from all resource providers, the auctioneer calculates the bid performance ratio (BPR) for each task and then determines the winner for each task. The bid of resource provider j∈J is defined as the weighted ratio of the bid price to the resource provider's processing capacity and the ratio of the data size to the transmission rate. m,j To express the bid performance ratio (BPR), as follows:

[0098]

[0099] Among them, δ∈[0,1] is a control coefficient, and the busy terminal can adjust it according to whether it needs more powerful processing power or more stable transmission. According to the definition of bid performance ratio (BPR), busy terminals want to get the maximum performance improvement with the least cost. Therefore, resource providers with lower bid performance ratio (BPR) should have a higher chance of winning the bid, in order to meet the service quality requirements of busy terminals.

[0100] Step S4: Task scheduling sequence scheme based on urgency-profit ratio

[0101] The matching process of tasks and resource providers works in a round-by-round manner to determine the matches between tasks and resource providers. The matching process of each round is as follows. Each computing task m∈M is first matched with the resource provider with the lowest bid-performance ratio (BPR). There are two situations:

[0102] (1) The number of tasks matched by resource provider j∈J is less than or equal to the maximum number of tasks it can handle simultaneously. In this case, the utility of the system completing the tasks is calculated as follows:

[0103]

[0104] For U m,j Tasks with a value less than 0 will be discarded. m,j Tasks ≥ 0 are added to the matching task set M assigned Then update the waiting delay and bidding price on resource provider j∈J for the next auction round. Then the current auction (the kth auction) Will be updated to look like this:

[0105]

[0106] in, represents the set of tasks assigned in the kth round of auction, and then according to Update bid priceb m,j (k+1). Then the corresponding bidding performance ratio γ m,j The values ​​will also be updated to reflect changes in wait delays and bid prices.

[0107] (2) The number of tasks matched by resource provider j∈J is greater than the maximum number of tasks n that it can process simultaneously j In this case, tasks are divided according to the profit-urgent ratio λ m,j Arranged from large to small, resource provider j∈J will select the first n j tasks, if in the first n j tasks have a profit-urgent ratio λ m,j Tasks less than 0 will be directly discarded to maximize system utility. m,j , which is defined in the present invention as the ratio of system utility to time urgency, expressed as follows:

[0108]

[0109] Then add the corresponding task to the set M assigned In, and according to the updated M assigned To update the waiting delay of resource provider j∈J for the next auction Bid price m,j (k+1) and bidding performance ratio γ m,j value, then all remaining tasks matching resource provider j∈J in this round of auction will be moved to set M unassigned After all resource providers are matched, if The next round of auction will be conducted until each task is matched with a resource provider, and then the auction will stop.

[0110] Step S5: Send the optimal result obtained in step S4 to the task executor;

[0111] Step S6: After the task is completed, the resource provider charges the corresponding buyer and returns its calculation results.

[0112] Furthermore, the present invention evaluates the performance of the method proposed in the present invention through simulation experiments and comparison with different algorithms, including an offloading algorithm that only offloads tasks to edge servers (edge ​​cloud offloading algorithm), an offloading algorithm with random processing order (random scheduling, RS), and a first-in-first-out task processing order offloading algorithm (FIFO). The method of the present invention is hereinafter referred to as BMRA-TAS.

[0113] The simulation scene settings are as follows:

[0114] Set up a WiFi-based wireless access point and a corresponding edge computing node with a coverage radius of 350m. Set the number of busy terminals to range from 10 to 200, and the number of idle terminals to range from 30 to 90, all evenly distributed within the coverage area. The maximum D2D transmission radius is R d2d ≤25m. The computing power of the idle terminal is set to f j =[1.5,3.5]GHz, the computing power of the edge node is f0=7GHz, and the price of unit delay and unit energy consumption is set to [0.25,0.5]$ and 0.3$ respectively. The channel gain between the busy terminal and the resource provider is set to Where δ = 4 represents the path loss factor, D m,j It represents the distance between the busy terminal m and the resource provider j, and the transmission power is 1W, the total transmission bandwidth B=40MHZ, and the noise power spectrum density is -174dBm / Hz.

[0115] For the computing tasks generated by the busy terminal m∈M, the task valuation is set to v in the simulation. m =[0.1,10]$, the data size is set to d m =[5,10]MB, maximum allowable delay τ m =[0.8,1]s, required computing resources C m =[0.5,1.2]Gcycles.

[0116] Figure 5 The relationship between the running time of the BMRA-TAS method proposed in the present invention and the number of busy terminals is demonstrated. Obviously, as the number of tasks increases, the running time gradually increases. When the number of tasks is large, the running time is only about 0.7s, which will not have much impact on the execution of the task and will not affect the real-time requirements of unloading tasks in the MEC system.

[0117] Figure 6 , 7 The trend of system utility and the number of effectively processed tasks as the number of task requests and the number of resource providers changes is shown. Specifically, Figure 6 In , the system utility increases with the increase in the number of tasks, but the utility at CoL = 90 is significantly higher than that at CoL = 70 and CoL = 50. This shows that more idle terminals can improve the system processing efficiency and meet the delay requirements of more tasks, thereby reducing the penalty cost of the system due to timeout completion of tasks, and thus improving the overall utility of the system. However, as the amount of tasks increases to a certain value, the system utility gradually tends to be flat. This is because when the unloaded tasks exceed the system processing capacity, the timeout penalty cost increases, the system utility growth rate slows down, and at the same time, the tasks with negative system utility will be discarded, so the system utility gradually tends to be stable. Figure 7In the figure, as the number of busy terminals and idle terminals increases, the number of effective service task requests also increases first and then tends to stabilize. This is because the system computing resources are limited. As the number of task requests increases, the system's maximum processing capacity will be reached. When CoL=70, the number of effective requests is significantly higher than CoL=50 and CoL=30, because more idle terminals can better handle the requests generated by a large number of busy terminals, share the burden of processing tasks, and improve the overall system's service capabilities.

[0118] Figure 8 , Fig. 9 By comparing with different algorithms, the performance of the BMRA-TAS algorithm proposed in the present invention is evaluated (CoL=70). Figure 8 It can be seen that the BMRA-TAS method proposed in the present invention is superior to the edge cloud offloading algorithm, RS offloading algorithm and FIFO offloading algorithm. In addition, the BMRA-TAS curve has been in an upward state, while for the RS offloading algorithm and the FIFO offloading algorithm, since both algorithms cannot guarantee the rationality of the participants, as the penalty cost and task processing cost caused by the task violating the delay requirement increase, the curve will first rise and then fall rapidly. The curve of the edge cloud offloading algorithm quickly stabilizes after rising. This is because the edge cloud computing resources are limited. As the number of tasks increases, the processing limit of the edge cloud is quickly reached, and the tasks that exceed the edge cloud processing capacity are directly discarded, resulting in no profit for the system, so the curve tends to be stable. Fig. 9 It can be seen that under the control of the BMRA-TAS method proposed in the present invention, the number of effectively processed task requests (i.e., tasks processed within the allowed delay range) is greater than that of the edge cloud offloading algorithm, RS offloading algorithm, and FIFO offloading algorithm. This proves that under the control of BMRA-TAS, the system can increase the number of effectively processed task requests while improving the system utility by adjusting the execution order of computing tasks.

[0119] What is disclosed above is only a preferred embodiment of the present invention, and it certainly cannot be used to limit the scope of rights of the present invention. Ordinary technicians in this field can understand that all or part of the processes of the above embodiment and equivalent changes made according to the claims of the present invention still fall within the scope of the invention.

Claims

1. A task allocation and scheduling method based on multi-round auctions, characterized in that: The following steps are involved: Step 1: Build an auction framework including auctioneers, buyers and sellers; Step 2: After receiving the task information published by the auctioneer, the resource provider submits a task bid to the auctioneer; Step 3: The auctioneer selects the winner of each task based on the bid-performance ratio of each task; Step 4: Task scheduling sequence plan based on urgency-profit ratio; Step 5: Send the optimal result obtained in step 4 to the task executor; Step 6: After the task is completed, the resource provider charges the corresponding buyer and returns its calculation results.

2. The task allocation and scheduling method based on multi-round auctions as claimed in claim 1, characterized in that: The auctioneer in step 1 is a trusted third party, i.e., a central cloud, an edge cloud, or a professional auction organization. The buyer is a busy terminal, and the seller is an idle terminal. All participants are honest during the auction process.

3. The task allocation and scheduling method based on multi-round auctions as claimed in claim 2, characterized in that: During the execution of step 2, the auctioneer sends the information of each task (d m ,C m ,t m ,τ m ) is sent to resource providers for auction, where d m is the amount of data for the computing task m∈M, C m is the computing resources required to process the task, τ m is the maximum allowable delay of task m∈M, t m is the time when task m∈M is generated on a busy terminal, M={1,2,...,M} represents the set of busy terminals, C={1,2,...,j,...,C} represents the set of all idle terminals; After receiving the task information, the resource provider submits the task bid to the auctioneer. They represent the bid price p of resource provider j∈J for task m∈M in the kth auction round. m,j (k), the computing power provided f j , and the waiting delay required for the task J = {C∪0} represents the set of resource providers for all tasks in the system, and index 0 represents the edge server.

4. The task allocation and scheduling method based on multi-round auctions as claimed in claim 3, characterized in that: The bidding performance ratio expression in step 3 is as follows: Among them, δ∈[0,1] is the control coefficient.

5. The task allocation and scheduling method based on multi-round auctions as claimed in claim 4, characterized in that: In step 4, each computing task m∈M is first matched with the resource provider with the lowest bid performance ratio. There are two situations: The number of tasks matched by resource provider j∈J is less than or equal to the maximum number of tasks it can process simultaneously; The number of tasks matched by resource provider j∈J is greater than the maximum number of tasks n that it can process simultaneously j .

6. The task allocation and scheduling method based on multi-round auctions as claimed in claim 5, characterized in that: When the number of matching tasks is less than or equal to the maximum number of tasks that the resource provider can handle simultaneously, the utility of the system completing the tasks is calculated as follows: Among them, v m is the true estimate of the task, p e,j , p t,j They represent the economic cost of unit energy consumption and delay respectively; For U m,j The task with U<0 will be discarded. m,j Tasks ≥ 0 are added to the matching task set M assigned Then update the waiting delay and bid price on resource provider j∈J for the next round of auction. Will be updated to look like this: in, Represents the set of tasks assigned in the kth round of auction, and then according to Update bid priceb m,j (k+1), then the corresponding bidding performance ratio γ m,j The values ​​will also be updated to reflect changes in wait delays and bid prices.

7. The task allocation and scheduling method based on multi-round auctions as claimed in claim 5, characterized in that: When the number of matching tasks is greater than the maximum number of tasks that the resource provider can handle simultaneously, the tasks are divided according to the profit-urgent ratio λ. m,j Arranged from large to small, resource provider j∈J will select the first n j tasks, if in the first n j tasks have a profit-urgent ratio λ m,j Tasks with a value less than 0 will be discarded directly; Profit-Emergency Ratio λ m,j is the ratio of system utility to time urgency, expressed as follows:

Images