Service unloading method and system based on multi-level unloading network
Through dynamic pricing of multi-level offload network and demand functions, devices are encouraged to participate in D2D calculation and unload, solving the problems of waste of resources and low utilization, and achieving efficient resource allocation and maximization of user benefits.
Patent Information
- Application Number
- CN202411023738.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-29
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-07-29
AI Technical Summary
In the prior art, the problems of waste of resources, low utilization of auxiliary computing equipment in the community and low returns of task requesters, especially in the D2D computing unloading scenario, the lack of effective incentive mechanisms leads to low equipment participation and low resource utilization.
Using the service unloading method of multi-level offloading network, through building network models, executing pricing algorithms for quantized resource models, and multi-level task unloading algorithm MOS, the demand function is introduced to dynamic pricing resources, and the equipment is encouraged to participate in task unloading, and optimized through market-oriented mechanisms. Resource allocation ensures maximum benefits for equipment and users.
It improves the utilization rate of auxiliary computing devices and the benefits of users, realizes efficient allocation of resources, reduces user costs, and shows good scalability, solving the problems of waste of resources and low utilization.
Smart Images

Figure CN119011659A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of service computing and D2D (Device-to-Device) communication, and in particular to a service offloading method and system based on a multi-level offloading network. Background Art
[0002] As the communication and computing capabilities of smart devices become increasingly powerful, more and more people need to use smart devices to obtain network information and process tasks. The rapid growth of traffic has brought many difficulties to the communication system, and edge servers are facing huge challenges such as overload and lack of spectrum resources. However, D2D can effectively solve these problems. Therefore, for the future sixth-generation (6G) mobile network, more and more studies are trying to introduce the mobile terminal layer into the cloud-edge two-layer computing paradigm. D2D can help edge servers respond to user requests, such as task computing requests, resource requests, etc., thereby alleviating the huge pressure on edge servers.
[0003] In the D2D computing offloading problem, the task requester transmits its own task to an idle device with sufficient resources (called an auxiliary computing device) through a direct device connection to request computing assistance. The auxiliary computing device completes the task calculation through its own resources. Since most devices are private and selfish, how to motivate devices to participate in offloading is a widely studied issue. If there is a lack of sufficient motivation and incentives, the auxiliary computing device is likely to avoid participating in the offloading network, resulting in a waste of resources. In addition, due to the limitation of the device communication range, the auxiliary computing devices around the task requester only account for a part of all the auxiliary computing devices in the community (such as a residential area, an industrial area). Therefore, if a single-hop direct connection is adopted, it is bound to cause a low utilization rate of the auxiliary computing devices in the community. Therefore, it is necessary to design a computing offloading mode to motivate more auxiliary computing devices to participate in task offloading, improve the utilization rate of idle devices, and optimize the benefits of the task requester.
[0004] The existing invention patent application document "A D2D task offloading system and its multi-type resource trading method" with publication number CN111343595A, the existing method includes: first, mobile users are grouped according to the connected components of the network topology to improve the efficiency of resource allocation. Then, the resources are priced using McAfee bilateral auctions, and finally, the mobile users are divided into multiple virtual traders, and the final resource allocation plan is obtained using the maximum matching method. From the specific implementation content of the existing plan, it can be seen that each mobile user reports his demand / supply of resources and the bid / ask price for each resource. The goal of the existing plan is to maximize the number of successful service transactions through auctions, and the final pricing for multiple resources is jointly determined by the bids / asks of all buyers and sellers. The number of successful transactions is maximized by determining the final resource transaction price. This auction method requires a server or other device to act as an auctioneer, which is a central control node with global information. The model adopted by the existing plan has the risk of single point failure.
[0005] The existing invention patent application document "A MEC pricing and unloading method based on D2D communication in the power Internet of Things" with publication number CN111107153A includes: completing the uniform distribution of tasks within the team through D2D data communication technology; each pair of paired devices determines its own data unloading amount; determines a pricing set and sorts the pricing set; the MEC server broadcasts a pricing plan in the pricing set in sequence; the inspection device determines its own unloading plan and uploads it to the MEC server; the cloud server determines its own revenue based on the calculation plan of the inspection device, and determines whether the pricing plan needs to be updated according to the principle of maximizing revenue; the cloud server outputs the optimal pricing plan through iteration. The focus of this existing solution is also on resource pricing, and resource pricing directly affects the task unloading decision of the device. In addition, this existing solution determines pricing and unloading decisions through game theory, and prices and unloading decisions are jointly affected by resource providers and requesters.
[0006] The existing public document "Research on D2D-assisted Computation Offloading Strategy in Mobile Edge Computing Network" states that since the computing and communication resources in the network are always limited, reasonable resource allocation according to the network conditions can improve the overall performance of the computing offloading system and ensure the service quality of all users. Resource allocation in the computing offloading process is to allocate server computing resources and network communication resources with the goal of minimizing task completion delay, completion energy consumption, or the trade-off between delay and energy consumption. In the problem of computing resource allocation, usually under the premise of limited server computing resources, it is necessary to find a reasonable solution to allocate limited computing resources to all users so that the objective function is optimized. The task offloading scheme recorded in the existing literature adopts a combination of D2D and edge servers, which requires the edge server to play a role and regards the edge server as a central control node with a global perspective; in terms of incentives for users, the existing literature adopts social influence and profit incentive mechanisms to encourage devices to participate in offloading, that is, the current participation in offloading will affect the subsequent offloading actions of the device; the existing literature studies the multi-requester scenario and does not consider the application scenario of single user task offloading; the existing literature adopts a game approach to solve the problem, but the game is more complex and the algorithm execution time is longer, especially in the application scenario of expanding the system scale.
[0007] In summary, the existing technology has technical problems such as waste of resources, low utilization rate of auxiliary computing equipment in the community, and low benefits for task requesters. Summary of the invention
[0008] The technical problem to be solved by the present invention is: how to solve the technical problems of resource waste, low utilization rate of auxiliary computing equipment in the community and low benefits for task requesters in the prior art.
[0009] The present invention adopts the following technical solution to solve the above technical problem: a service unloading method based on a multi-level unloading network comprises:
[0010] S1. Build a network model, where the devices in the network model include: user equipment and service provider equipment;
[0011] S2. Perform abstract operations on the network model and use preset quadruple to describe the devices in the network model and the task sets generated by the devices:
[0012] S3. Execute the pricing algorithm of the quantitative resource model, wherein the demand function is obtained and based on the transaction history information, the request price and the demand rate of n resources are calculated to fit the price and demand relationship of a single resource, and the resource pricing problem is converted into the problem of maximizing the expected return of the equipment to obtain the optimal return of the equipment and the optimal price of the resource;
[0013] S4. Taking the user device as the center and the maximum task completion delay as the constraint, a multi-level task offloading network is constructed to execute the task offloading algorithm MOS of the multi-level offloading network to perform layer-by-layer quotation operations and obtain applicable task offloading solutions.
[0014] The present invention is based on a multi-level offloading network and introduces a demand function to make service pricing follow demand changes, and designs a D2D service offloading method. The computing offloading mode designed by the present invention can encourage more auxiliary computing devices to participate in task offloading, improve the utilization rate of idle devices, and optimize the benefits of task requesters.
[0015] The goal of the present invention is to maximize the benefits of user resource offloading. Resource pricing is determined only by the device itself and is related to the remaining amount of resources and remaining selling time of the device itself. Therefore, the resource price is not affected by other devices and will not change once determined.
[0016] In a more specific technical solution, S2 includes:
[0017] S21, perform a device definition operation, wherein a preset four-tuple (d, s, p, r) is used to describe the device, wherein d is the device number; s represents the connection vector between devices; p represents the price list; and r represents the task set;
[0018] S22. Perform a task set definition operation. The task set consists of no less than 2 subtasks. The task set is described using a preset quadruple (d, n, q, td), where d represents the unique identifier of the task set, n represents the size of the task set, q is a resource list, and td represents the deadline for completion of the task set.
[0019] In more specific technical solutions, S3 includes:
[0020] S31. Use the following logic to correspond the task set to the request type vector:
[0021]
[0022] Indicates the number of resources 1, 2, ..., n required by the tasks in the task set S at each moment;
[0023] S32. According to the transaction history information, the demand function is obtained:
[0024] σ=Ω(P)
[0025] Where σ and P represent the request arrival rate and service price respectively;
[0026] S33. Calculate the requested price using the following logic:
[0027]
[0028] In the formula, q τ,λ represents the price of resource λ in time slice τ;
[0029] S34. Determine the demand rates of n resources according to the request type vector, the demand function and the request price using the following logic:
[0030]
[0031] In the formula, (S 1 ,S 2 ,...,S m ) represents the m types of services deployed by the device, is the arrival rate of request i;
[0032] S35. According to the request type vector, the demand function, the request price, and the demand rate of the n resources, a single resource price and demand relationship is obtained by fitting:
[0033] φ λ =Θ λ (q)
[0034] S36. Formalize the resource pricing problem of equipment;
[0035] S37. Use the following logic to determine the optimal benefit of the equipment:
[0036]
[0037] S38. Determine the optimal price of resources using the following logic:
[0038] q τ,λ =V * (λ,t,n)-V * (λ,t,n-1)+1.
[0040] The present invention proposes a resource pricing scheme for equipment, and proposes that in D2D communication, the user maximizes the profit when performing task offloading. The present invention designs a multi-level task offloading algorithm MOS, so that the user's profit from completing the task is maximized.
[0041] The present invention integrates the concept of marketization into the service provision framework to price the resources of the device. The interconnection between devices builds an offloading network, and users look for the best resource purchase solution in the offloading network to maximize profits.
[0042] In a more specific technical solution, S36 includes:
[0043] S361. Use the following logic to express the expected benefits of the device:
[0044]
[0045] S362. Use the following logic to transform the resource pricing problem into the problem of maximizing the expected benefits of the equipment:
[0046]
[0047] S363, using the following demand function logic, the demand for resource λ is modeled as following a Poisson distribution with an intensity of Φ λ A random process:
[0048]
[0049] Where k and α are constants.
[0050] The present invention formulates a service offloading algorithm so that users can purchase resources of idle devices to complete tasks. A pricing function based on the resource status of the device is formulated so that all devices can obtain satisfactory benefits.
[0051] In a more specific technical solution, the task offloading algorithm MOS of the multi-level offloading network in S4 includes:
[0052] S41, constructing a multi-level offloading network, wherein, from the devices, selecting a suitable auxiliary computing device; identifying the predecessor device and the successor device of the auxiliary computing device to determine the task transfer direction;
[0053] S42, a user benefit function is formulated to represent the benefit of each intermediate auxiliary computing device, to determine the price that the intermediate auxiliary device should pay for purchasing resources from the subsequent device, and to convert the goal of maximizing user benefits into the goal of maximizing the benefit of each device itself, so as to minimize its own cost;
[0054] S43. Execute a multi-level offloading strategy, sort the pricing of the intermediate auxiliary computing devices, determine and make an offloading task allocation decision based on the resource selling price in the bid table.
[0055] In a more specific technical solution, S41 includes:
[0056] S411. Determine the execution time and transmission time of the total task according to the task size C:
[0057] t e +t i,j ≤τ j
[0058]
[0059] Where, t e is the total task execution time, ti,j is the transmission time, τ j is the delay constraint of the task received by device j, I and O are the input and output sizes of a single task, r is the transmission rate, and j satisfies the delay constraint;
[0060] S412, the current device i subtracts t from the delay constraint in the message i,j , and increase the hop count by 1, and send the message to the next device j;
[0061] S413: When all successor devices of the current device i cannot meet the delay constraint, it is determined that the current device i is the last layer node of the multi-level offloading network.
[0062] The multi-level task offloading algorithm provided by the present invention is user-centric and constructs a multi-level task offloading network with the maximum task completion delay as a constraint. The user decides the final task offloading solution through layer-by-layer quotation.
[0063] In a more specific technical solution, S42 includes:
[0064] S421. Use the following logic to formulate a user benefit function:
[0065]
[0066] In the formula, It indicates the value of resources consumed by the user to complete the task alone, and Cost' indicates the value of the resources actually consumed by the user to complete the task. It is the incentive given to device j when the user offloads the task to device j, and V represents the benefit that the user can obtain by completing the task. The present invention studies the scenario of task offloading by a single user.
[0067] S422, obtaining the expected rate of return ERR of each intermediate auxiliary computing device. When the intermediate auxiliary computing device receives the bid list of the successor node, it multiplies the bid list by the expected rate of return ERR of the current intermediate auxiliary computing device, and formulates the bid list of the current intermediate auxiliary computing device according to the resource pricing table and the expected rate of return; wherein the income of each intermediate auxiliary computing device is determined by the following logic:
[0068] W i =R i -Q i
[0069] The revenue from the last layer of auxiliary computing equipment in the multi-level offloading network is used as the resource selling price:
[0070] W′=Q′
[0071] S423: Determine the reward R given by the predecessor node to device i according to the bid table of the predecessor node according to device i. i , based on which the following logic is used to determine the price Q that the intermediate auxiliary computing device should pay for purchasing resources from the subsequent device i :
[0072]
[0073] In the formula, represents the sum of all resource prices of the assigned task completed by device j, q λ,τ represents the price of resource λ in time period τ, o is the number of all resources, Su(i) is the set of successor devices of device i, and Se(j) is the set of time slices for choosing to purchase the resources of device j.
[0074] S424. Using the following logic, the goal of maximizing user benefits is converted into the goal of maximizing each device's own benefits, so as to minimize its own costs:
[0075] Q i
[0076] stj∈Su(i)
[0077]
[0078] Where C is the size of the total task, is the amount of tasks executed locally on device i, C i is the amount of tasks received by device i.
[0079] The present invention only uses a profit mechanism to motivate devices to participate in uninstallation. The uninstallation behavior is one-time and will not affect subsequent uninstallation behaviors.
[0080] In a more specific technical solution, in S43, the operation of arranging the price list includes:
[0081] S431, within the preset time constraint range, according to the type and quantity of resources required by a task in each time slice, calculate the time slice execution task price;
[0082] S432, according to the time slice with the lowest price for executing the task, select the time slice with the lowest price, calculate the number of tasks m that can be processed in the minimum time slice, fill the price p required for the minimum price time slice into the price table, add the price p to the previous row price, and obtain the price of the min(m-1,n-1) row; remove the current minimum price time slice from the preset time constraint range;
[0083] S433. Repeat step S432 until the price list is filled with n rows, where n represents the number of tasks.
[0084] In a more specific technical solution, in S43, the operation of calculating the bid table includes:
[0085] S431', receiving bid lists from all subsequent devices;
[0086] S432', multiply the bid table value received by the subsequent device by the expected rate of return ERR of the intermediate auxiliary calculation device;
[0087] S433', set |Su(i)|+1 pointer p k , so that p k =1, k∈Su(i). Initialize variables S=0, l=1;
[0088] S434', according to the value pointed to by the pointer of the current table row |Su(i)|+1, select the smallest y, and record the table number corresponding to the a table, and use the following logic to perform the update operation:
[0089] S = S + y, T'(l) = S;
[0090] S435', update the bid table T a Values:
[0091] T a (m) = T a (m)-y,m∈{p a +1,p a +2,...,n}
[0092] Update Table T a Pointer and task number: p = p + 1, l = l + 1;
[0093] S436', repeat the above steps S434', S435' until l>n.
[0094] The present invention generates an offloading network by interconnecting devices, which allows users to complete tasks at a lower cost and improves the resource utilization of devices in a larger range. The present invention also measures its scalability and finds that the algorithm exhibits good scalability as the system scale increases.
[0095] The scenario of the present invention is a fully distributed scenario, which does not require the participation of a server. The device only needs and can only obtain the status information of the device directly connected to it, and expands the unloading network through the interconnection and communication between devices, reducing the user's unloading cost, which is the focus of the present invention.
[0096] In a more specific technical solution, the service offloading system based on the multi-level offloading network includes:
[0097] A network model building module, used to build a network model, wherein the devices in the network model include: user equipment and service provider equipment;
[0098] The network device description module is used to abstract the network model and use the preset quadruple to describe the devices in the network model and the task set generated by the devices. The network device description module is connected to the network model construction module:
[0099] The pricing algorithm module is used to execute the pricing algorithm of the quantitative resource model, wherein the demand function is obtained and based on the transaction history information, the request price and the demand rate of n resources are calculated to fit the price and demand relationship of a single resource, and the resource pricing problem is converted into the problem of maximizing the expected return of the equipment, so as to obtain the optimal return of the equipment and the optimal price of the resource;
[0100] The task offloading scheme acquisition module is used to build a multi-level task offloading network with user equipment as the center and the maximum task completion delay as the constraint, so as to execute the task offloading algorithm MOS of the multi-level offloading network, perform layer-by-layer quotation operations, and obtain applicable task offloading schemes. The task offloading scheme acquisition module is connected with the pricing algorithm module and the network device description module.
[0101] Compared with the prior art, the present invention has the following advantages:
[0102] The present invention is based on a multi-level offloading network and introduces a demand function to make service pricing follow demand changes, and designs a D2D service offloading method. The computing offloading mode designed by the present invention can encourage more auxiliary computing devices to participate in task offloading, improve the utilization rate of idle devices, and optimize the benefits of task requesters.
[0103] The present invention proposes a resource pricing scheme for equipment, and proposes that in D2D communication, the user maximizes the profit when performing task offloading. The present invention designs a multi-level task offloading algorithm MOS, so that the user's profit from completing the task is maximized.
[0104] The present invention integrates the concept of marketization into the service provision framework to price the resources of the device. The interconnection between devices builds an offloading network, and users look for the best resource purchase solution in the offloading network to maximize profits.
[0105] The present invention formulates a service offloading algorithm so that users can purchase resources of idle devices to complete tasks. A pricing function based on the resource status of the device is formulated so that all devices can obtain satisfactory benefits.
[0106] The multi-level task offloading algorithm provided by the present invention is user-centric and constructs a multi-level task offloading network with the maximum task completion delay as a constraint. The user decides the final task offloading solution through layer-by-layer quotation.
[0107] The present invention generates an offloading network by interconnecting devices, which allows users to complete tasks at a lower cost and improves the resource utilization of devices in a larger range. The present invention also measures its scalability and finds that the algorithm exhibits good scalability as the system scale increases.
[0108] The present invention solves the technical problems existing in the prior art of waste of resources, low utilization rate of auxiliary computing equipment in the community and low benefits for task requesters. BRIEF DESCRIPTION OF THE DRAWINGS
[0109] Figure 1 This is a schematic diagram of basic steps of a service offloading method based on a multi-level offloading network according to Embodiment 1 of the present invention;
[0110] Figure 2 This is a schematic diagram of an application scenario of a service offloading system based on a multi-level offloading network according to Embodiment 1 of the present invention;
[0111] Figure 3 This is a schematic diagram of specific steps of the task offloading algorithm MOS of the multi-level offloading network of Example 1 of the present invention;
[0112] Figure 4 This is a schematic diagram of specific steps for arranging a price list according to Embodiment 1 of the present invention;
[0113] Figure 5 This is an example diagram of the pricing table arrangement of Example 1 of the present invention;
[0114] Figure 6 Schematic diagram of specific steps for calculating a bid table according to Embodiment 1 of the present invention;
[0115] Figure 7 This is an example diagram of user task offloading according to Embodiment 2 of the present invention;
[0116] Figure 8 A first schematic diagram of effectiveness evaluation of an offloading algorithm according to Embodiment 2 of the present invention;
[0117] Fig. 9 This is a second schematic diagram of effectiveness evaluation of the offloading algorithm according to Embodiment 2 of the present invention. DETAILED DESCRIPTION
[0118] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0119] Example 1
[0120] like Figure 1 and Figure 2 As shown, the service unloading method based on the multi-level unloading network provided by the present invention includes the following basic steps:
[0121] S1. Build a network model. The roles in the network model include but are not limited to: users and service providers;
[0122] In this embodiment, the roles involved in the network model are:
[0123] User: Task owner and producer, whose mobile device resources are limited and insufficient to complete the task. They will purchase resources from other devices to complete the task based on the offloading decision algorithm.
[0124] Service Provider: A mobile device that has sufficient resources to maintain its own equipment and can earn revenue by selling its own resources.
[0125] S2. Perform abstract operations on the network model:
[0126] In this embodiment, a device definition operation is performed, where both the user and the service provider are devices. Each device can be described by a four-tuple (d, s, p, r), where the meaning of each element is:
[0127] (1) Serial number of the equipment;
[0128] (2) s connection vector, indicating the connectivity status between the device and other devices;
[0129] (3) p price table, indicating the price of each resource unit of the equipment in the future;
[0130] (4) rTask set. All devices may generate a task set. The device that generates the task set becomes the user and needs to purchase resources from other service provider devices to complete the task set.
[0131] In this embodiment, a task set definition operation is performed: a task set is composed of multiple subtasks, and each subtask is equal. A task set can be described by a four-tuple (d, n, q, td), where the meaning of each element is:
[0132] (1) d is the unique identifier of the task set, which is used to identify the user who generated the task;
[0133] (2) n represents the size of the task set;
[0134] (3) q is a resource list, indicating the number of different types of resources required by each subtask in each time slice;
[0135] (4) td represents the deadline for completion of the task set.
[0136] S3. Execute the pricing algorithm of the quantitative resource model, wherein the demand function is introduced, the request price and the demand rate of n resources are calculated, the relationship between the price and demand of a single resource is fitted, and the resource pricing problem is transformed into the problem of maximizing the expected benefit of the device, and the optimal benefit of the device and the optimal price of the resource are obtained;
[0137] In this embodiment, the auxiliary computing devices in the offloading network have sufficient resources, and selling resources is the main way to earn revenue. In this embodiment, the value of resources is determined by the remaining amount of resources and the remaining time of selling resources. Since the task set proposed by the user contains multiple identical tasks, a task set can be mapped to a request type, vector:
[0138]
[0139] It represents the number of resources 1, 2, ..., n required by the tasks in the task set S at each moment. For price request, in the pricing algorithm of the quantitative resource model in step S3, the concept of demand function in economics is introduced;
[0140] In this embodiment, the aforementioned demand function can be obtained from the transaction history:
[0141] σ=Ω(P)
[0142] Where σ and P represent the request arrival rate and service price respectively. Thus, the request price is expressed as follows:
[0143]
[0144] Among them, q τ,λ represents the price of resource λ in time slice τ. Combining the above formula, the demand rate of n resources is expressed as follows:
[0145]
[0146] Among them, (S 11 ,S 2 ,...,S m ) is the m type of service deployed by the device, is the arrival rate of request i.
[0147] Combining the above four formulas, we can fit the relationship between the price and demand of a single resource:
[0148] Φ λ =Θ λ (q)
[0149] After the resource unit price and demand relationship are determined, the resource pricing problem of the equipment will be formalized.
[0150] In this embodiment, for a given time t and resource type λ, n=ξ λ,t Indicates that at time t, there are n units of idle resources λ on the device, n k,λ represents the quantity of resource λ sold at time k, q τ,λ Indicates the price of resource λ at time τ. In this embodiment, represents the quantity of resource λ sold until time τ. Therefore, the expected revenue of the equipment can be expressed as follows:
[0151]
[0152] In this embodiment, the resource pricing problem is transformed into the problem of maximizing the expected revenue of the device:
[0153]
[0154] Subsequently, the demand for resource λ is modeled as a Poisson distribution with intensity Φ λ Using the classic demand function:
[0155]
[0156] Wherein, k and α are both constants, and have no effect on the trend of price changes over time and the amount of available resources. For the convenience of representation, in this embodiment, α=1 is set, and the optimal benefit of the equipment can be expressed as follows:
[0157]
[0158] The optimal price of resources can be calculated by the following formula:
[0159] q τ,λ =V * (λ,t,n)-V * (λ,t,n-1)+1
[0160] It can be seen that less remaining resource quantity or more remaining selling time will make the price of the resource higher.
[0161] S4. Using the task offloading algorithm MOS of the multi-level offloading network, a multi-level task offloading network is built with users as the center and the maximum task completion delay as the constraint. Through the layer-by-layer quotation method, users can decide the final task offloading plan;
[0162] like Figure 3 As shown, in this embodiment, the task offloading algorithm MOS of the multi-level offloading network also includes the following specific steps:
[0163] S41, building an offloading network;
[0164] In this embodiment, since the task set has a maximum tolerable delay, the auxiliary computing device suitable for participating in the task offloading must be carefully selected. In addition, in order to ensure the smooth and efficient execution of the offloading task, the predecessor device and successor device of each device in the offloading network must be identified, that is, the direction of task transfer must be clear, which is crucial to ensure the coordination and orderliness of the offloading process. In order to maintain the accuracy of the bid list received by each device from its successor device, it must be ensured that if a device has multiple predecessor devices, it only selects one of them to communicate with it for task transfer. This approach helps prevent conflicts and inaccuracies in the bidding process and ensures the effectiveness of the offloading strategy.
[0165] In this embodiment, when a user generates a task set that needs to be offloaded, he will broadcast it to the devices connected to him that complete all tasks and meet the delay requirements. The broadcast includes the number of tasks, task tolerance delay, and number of hops. The device receiving the message will calculate which devices connected to it meet the requirements - whether the time to complete all tasks plus the time for task transmission between each other meets the experimental constraints. Since the offloading decision has not been determined at this time, C is used to represent the task size, that is:
[0166] t e +t i,j ≤τ j
[0167]
[0168] Among them, t e is the total task execution time, t i,j is the transmission time, τ j is the delay constraint of the task received by device j, I and O are the input and output size of a single task, and r is the transmission rate. If j satisfies the delay constraint, device i will subtract t from the delay constraint in the message. i,jAfter adding 1 to the hop count, the message is sent to device j, and device j makes the same choice. If all subsequent devices of device i cannot meet the delay constraint, then device i is the last node of the offload network.
[0169] S42. A user benefit function is formulated to represent the benefit of each intermediate auxiliary computing device (not the last layer of the network), determine the price that the device should pay to purchase resources from its successor device, and transform the goal of maximizing user benefits into maximizing the benefits of each device and minimizing its own costs;
[0170] In this embodiment, the user's perspective is considered to maximize the benefits of the user completing the task. For this purpose, the user's benefit function is formulated:
[0171]
[0172] in, It represents the value of resources consumed by the user to complete the task alone, and Cost' represents the value of the resources actually consumed by the user to complete the task. Therefore, the first two items represent the value of resources saved by the user when completing the task. The saved resources can still be sold, which is called the retained income of the user when completing the task. is the incentive given to device j when the user offloads the task to device j, V represents the benefit that the user can obtain by completing the task, and the last two items are called the real benefits of the user completing the task. Therefore, in order to maximize their own benefits, users will assign as many tasks as possible to devices with lower bids, pay lower rewards and retain their own higher-value resources.
[0173] In this embodiment, since the devices are selfish, each device wants to maximize its own benefits. However, the devices do not know each other's resource information. If the device bids high, it is very likely that it will not be able to obtain tasks from the predecessor node, resulting in zero benefits for itself. Therefore, the bid of the device will not only consider the pricing of its own resources, but also the resource bids of other devices connected to it. Two concepts should be distinguished here, pricing table and bidding table. For the last layer of the unloading network, the pricing table is a price table made by pricing the device's own resources using the pricing method mentioned above, and the bidding table is a compilation of its own resource pricing table. Since the pricing of resources is related to the remaining time and amount of remaining resources, the revenue obtained by the device is not linearly related to the amount of resources sold. In order to make the bid the lowest, the device will successively search for the time slice combination with the lowest price for executing n tasks within the total task execution time range, and make a bidding table. For the middle layer nodes of the network (non-last layer nodes), the bidding table of the device is a price table jointly formulated by the device based on its own sorted resource pricing table and the bidding table of the successor device, and then the device sends the bidding table to the predecessor node.
[0174] At the same time, in order to avoid the situation where the device cannot earn any profit simply as a relay node, each device will have a coefficient, which is called the expected rate of return (ERR) in this article. After receiving the bid list of the successor node, the device will multiply it by its own ERR, and then formulate the bid list in combination with its own resource pricing table. This ensures that even if the device does not sell its resources, it can still earn income by participating in task transmission. This also encourages devices to actively expand the offloading network and find better offloading strategies for users. Here is an explanation of the selection method of the optimal predecessor device in the previous article: for every additional hop between the device and the user, there will be an additional relay to charge a relay fee, and the probability of it being assigned a task will decrease. The revenue of each intermediate auxiliary computing device (not the last layer of the network) can be expressed as:
[0175] W i =R i -Q i
[0176] In this embodiment, the revenue of the auxiliary computing device at the last layer of the network is the selling price of the resource:
[0177] W′=Q′
[0178] Among them, R i It is the reward given by the predecessor node to device i, which is determined by the predecessor node according to the bid list of device i. i is the price that a device should pay to purchase resources from its successor device, Q i It is expressed as follows:
[0179]
[0180] in, represents the sum of all resource prices of the assigned task completed by device j, q λ,τ represents the price of resource λ in time period τ, o is the number of all resources, Su(i) is the set of successor devices of device i, and Se(j) is the set of time slices for choosing to purchase the resources of device j.
[0181] In this embodiment, device i cannot control R i , in order to maximize revenue, device i will choose the offloading strategy with the lowest bid, that is, minimize Q i Therefore, each node minimizes the offloading cost, and the goal of maximizing user benefits can be transformed into each device maximizing its own benefits, that is, minimizing its own costs:
[0182] Q i
[0183] stj∈Su(i)
[0184]
[0185]
[0186] Where C is the total task size, is the amount of tasks executed locally on device i, C i is the amount of tasks received by device i.
[0187] S43, executing a multi-level offloading strategy, ranking the pricing of the devices, using a price list to reflect the price of resource sales, and making task allocation decisions;
[0188] In this embodiment, in order to solve the task allocation problem, two parts need to be solved: the determination of the bid table and the offloading decision of each device. In order to determine the bid table, this section adopts a bottom-up approach, starting from the last layer of the offloading network. Since the pricing of device resources depends on the remaining sales time and the remaining resource amount, the pricing does not increase linearly. Therefore, it is necessary to sort the pricing of the devices, so that it can be ensured that the bid table accurately reflects the price of the resource sale, thereby making more effective task allocation decisions.
[0189] like Figure 4 and Figure 5 As shown, in this embodiment, the specific steps of arranging the price list include:
[0190] S431, within the time constraint, according to the type and quantity of resources required for each time slice of a task, calculate the price required to execute a task in the corresponding time slice;
[0191] S432, select the time slice with the minimum price required to execute the task, calculate the number of tasks m that can be processed in the time slice, and fill the required price p into the sorted pricing table, which means that the price required to execute a task is p, and the price of the subsequent min (m-1, n-1) rows is equal to the price of the previous row plus p. Finally, remove the time slice from the selection range;
[0192] S433. Repeat step S432 until the sorted price list is filled with n rows, where n is the number of tasks.
[0193] In the price list after the above sorting steps, the price of task execution is monotonically unchanged as the amount of tasks increases. After all devices are sorted, we start from the second-to-last layer device of the unloading network and calculate its bid list;
[0194] like Figure 6 As shown, in this embodiment, the specific steps of calculating the bid table include:
[0195] S431', receiving bid lists from all subsequent devices;
[0196] In this embodiment, the bid table is denoted by T b ,b∈Su(i). T 0 It represents the pricing table sorted by the device itself, and T' represents the bidding table i of the device, which is initially empty.
[0197] S432', multiply the bid table value received by the subsequent device by the ERR value of the device itself;
[0198] S433', set |Su(i)|+1 pointer p k , so that p k =1, k∈Su(i). Initialize variables S=0, l=1.
[0199] In this embodiment, the value of the pointer represents the amount of tasks assigned to device k, and initially points to the first row of the table. S represents the total price of unloading the current amount of tasks, and l represents the number of the task currently being assigned.
[0200] S434', according to the value pointed to by the current |Su(i)|+1 table row pointer, choose the smallest one, y, and write down the a table
[0201] Corresponding table number, update S=S+y, T'(l)=S;
[0202] S435', update table T a The value of
[0203] In this embodiment, the following logic is used to update table T a Values:
[0204] T a (m) = T a (m)-y,m∈{p a +1,p a +2,...,n}
[0205] Update Table T a Pointer and task number: p=p+1,l=l+1.
[0206] S436', repeat the above steps S434', S435' until l>n.
[0207] After the above steps are completed, the table T’ is the bid table of device i. After all the penultimate-layer devices transmit their bid tables to their respective predecessor devices, the devices in the third-to-last layer of the offloading network start the same operation until the devices in the first layer of the offloading network report their bid tables to the user device. Note that in step 5, the bid table selected in step 4 will be modified, and in each subsequent selection, considerations are based on the modified value. The reason for doing this is to find the device with the smallest price difference when executing task k, that is, given that the first k - 1 task allocation decisions have been made, select the minimum value among the rewards required for all devices to execute task k. It is not difficult to see that this method is a decision based on the greedy strategy. The greedy strategy selects the optimal solution for the next step based on the current state, and most of the time it can only achieve a local optimum rather than a global optimum. However, this method can obtain the global optimum when applied to this problem, which is the reason for sorting the price table before calculating the device bid table. The sorted price table satisfies the following conditions:
[0208] and x, y > 0, when 0 < x < y < n, there is T a (y) - T a (y - 1) ≥ T a (y) - T a (y - 1)
[0209] ≥ T a (x) - T a (x - 1).
[0210] After the price table satisfies the foregoing conditions, each optimal decision made in the current state is a step closer to the global optimum. Thus, the first part of this problem, that is, the determination part of the bid table, is well solved.
[0211] As for the second part, the user offloading decision problem, it is the reverse process of the first part. To solve this part of the problem, starting from the user device and along the direction of increasing the offloading network layer depth, each device performs the following operations:
[0212] Set |Su(i)| + 1 pointers p k , such that p k = 1, k ∈ Su(i), initialize the variable l = 1, and the initial pointers all point to the first row of the table. l represents the task number currently being allocated; according to the values in the rows pointed to by the pointers in the current |Su(i)| + 1 tables, select the smallest one and denote it as |Su(i)| + 1, and at the same time record the corresponding table number a; update the value of table T a , T a (m) = T a (m) - y, m ∈ {p a + 1, pa +2,...,n}. Update table T a pointer and task number: p = p + 1, l = l + 1; repeat the above sub-steps until l>n', where n' is the number of tasks that device i receives from its predecessor device, and n' = n for user device; obtain the task allocation scheme according to the pointer pointing situation of each table, for example, if p a =3, then assign 2 tasks to the successor device a, and calculate its own profit based on this task allocation decision.
[0213] In this embodiment, each device must follow the steps from the user device to the last layer of the offload network to determine the number of tasks to be assigned to it. Once the task allocation decision is made, both the user and the participating devices can calculate their respective benefits.
[0214] Example 2
[0215] In this embodiment, the user offloading process is introduced by taking video processing as an example. After the user generates a task, he will calculate the cost required to complete the task according to the pricing method proposed in this article, and then calculate which devices connected to him can complete the task on time according to the deadline of the task, and send a message containing the task size, task deadline and number of hops to these devices through the D2D connection. These devices form the first layer of the offloading network. After the aforementioned devices return a confirmation message to the user, they repeat this action until the offloading network is built.
[0216] like Figure 7 As shown, in this embodiment, the last layer of the network device aggregates the resource prices within the video processing time into a price list and organizes it, and sends it to the upper layer device through a message. After receiving the price list of all subsequent devices, the upper layer device multiplies it by its own ERR, and then combines it with its own price list to select the time combination with the shortest resource price required to process the entire task, organizes it into a bid list, and then returns it to its own predecessor device. This report is reported layer by layer until the user collects the bid list of its subsequent devices, compares it with its own resource price list, formulates the final task offloading strategy, and then performs task offloading.
[0217] In this embodiment, based on the model provided by the present invention, we set up 22 devices, randomly selected one of them as a user, simulated the task offloading process, and set up other 3 comparison methods for control experiments:
[0218] FDS: This method is also a multi-device multi-layer task offloading method, which seeks the most user-friendly offloading solution by offloading layer by layer. However, this method does not consider the benefits of the participating offloading devices as relay devices. The method in this chapter makes up for this.
[0219] SOSRAS: This method selects one device from among many devices connected to the user's device to perform partial offloading so that all tasks can be completed in the shortest time. When the task is completed before the deadline, the user will have additional benefits, and the shorter the task completion time, the higher the additional benefits.
[0220] TAROC: This method is a double auction method. When a device directly connected to the user successfully bids for a task, it will auction the task again to earn revenue.
[0221] In this embodiment, in order to verify the effect of the method in this paper under the change of task volume, we change the task volume and record the user's benefits.
[0222] like Figure 8 As shown, in this embodiment, the user benefits obtained by the above four methods increase with the increase of the number of tasks. This is reasonable because each task will bring certain benefits to the user, and the more tasks the user has, the greater the total benefit. When the number of tasks is small, the difference between the above four methods is not large. However, as the number of tasks increases, the performance of the method in this paper is significantly better than the other three methods. This is because in the pricing method mentioned above, the price difference of resources in different time slices is not very large, and when the number of tasks is small, the number of time slices involved is small, so the difference in user benefits is not very obvious. As the number of tasks increases, there are more and more time slices available for users to choose in the entire unloading network. The results show that when the price differences of resources in different states are accumulated over multiple time slices, the difference in user benefits is significant.
[0223] like Fig. 9 As shown in FIG. 1 , in this embodiment, the effect of the number of devices in each layer on the method is investigated. Specifically, an offloading network with a network depth of 2 is created, and the number of subsequent devices is changed for each device in the network. Then, the changes in user income are recorded and Fig. 9 The experimental results are plotted in . As the number of subsequent devices increases, it can be observed that the growth rate of user revenue for the proposed method and the FDS method gradually slows down and eventually flattens out. This can be attributed to the fact that the range of options available to users increases with the number of subsequent devices. However, once the number of subsequent devices reaches a certain value, it is difficult to find a better allocation scheme, resulting in no significant change in user revenue. When users use the proposed method to uninstall, the revenue is significantly better than other methods.
[0224] In summary, the present invention is based on a multi-level offloading network and introduces a demand function to make service pricing follow demand changes, and designs a D2D service offloading method. The computing offloading mode designed by the present invention can encourage more auxiliary computing devices to participate in task offloading, improve the utilization rate of idle devices, and optimize the benefits of task requesters.
[0225] The present invention proposes a resource pricing scheme for equipment, and proposes that in D2D communication, the user maximizes the profit when performing task offloading. The present invention designs a multi-level task offloading algorithm MOS, so that the user's profit from completing the task is maximized.
[0226] The present invention integrates the concept of marketization into the service provision framework to price the resources of the device. The interconnection between devices builds an offloading network, and users look for the best resource purchase solution in the offloading network to maximize profits.
[0227] The present invention formulates a service offloading algorithm so that users can purchase resources of idle devices to complete tasks. A pricing function based on the resource status of the device is formulated so that all devices can obtain satisfactory benefits.
[0228] The multi-level task offloading algorithm provided by the present invention is user-centric and constructs a multi-level task offloading network with the maximum task completion delay as a constraint. The user decides the final task offloading solution through layer-by-layer quotation.
[0229] The present invention generates an offloading network by interconnecting devices, which allows users to complete tasks at a lower cost and improves the resource utilization of devices in a larger range. The present invention also measures its scalability and finds that the algorithm exhibits good scalability as the system scale increases.
[0230] The present invention solves the technical problems existing in the prior art of waste of resources, low utilization rate of auxiliary computing equipment in the community and low benefits for task requesters.
[0231] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A service unloading method based on a multi-level unloading network, characterized in that: The method comprises: S1. Constructing a network model, wherein the devices in the network model include: user equipment and service provider equipment; S2. Perform an abstract operation on the network model, and use a preset quadruple to describe the devices in the network model and the task set generated by the devices: S3. Execute the pricing algorithm of the quantitative resource model, wherein the demand function is obtained and based on the transaction history information, the request price and the demand rate of n resources are calculated to fit the price and demand relationship of a single resource, and the resource pricing problem is converted into the problem of maximizing the expected return of the equipment to obtain the optimal return of the equipment and the optimal price of the resource; S4. Taking the user equipment as the center and the maximum task completion delay as the constraint, a multi-level task offloading network is constructed to execute the task offloading algorithm MOS of the multi-level offloading network to perform layer-by-layer quotation operations and obtain an applicable task offloading solution.
2. The service unloading method based on a multi-level unloading network according to claim 1, characterized in that: The S2 includes: S21, performing a device definition operation, wherein the device is described using the preset quadruple (d, s, p, r), wherein d is the device number; s represents the connection vector between the devices; p represents the price list; and r represents the task set; S22. Perform a task set definition operation, where the task set includes no less than 2 subtasks, and the task set is described using the preset quadruple (d, n, q, td), wherein d represents the unique identifier of the task set, n represents the size of the task set, q is a resource list, and td represents the deadline for completion of the task set.
3. The service unloading method based on a multi-level unloading network according to claim 1, characterized in that: The S3 includes: S31. Use the following logic to correspond the task set to the request type vector: Indicates the number of resources 1, 2, ..., n required by the tasks in the task set S at each moment; S32. According to the transaction history information, the demand function is obtained by processing: σ=Ω(P) Where σ and P represent the request arrival rate and service price respectively; S33. Calculate the requested price using the following logic: In the formula, q τ,λ represents the price of resource λ in time slice τ; S34. Determine the demand rates of the n types of resources according to the request type vector, the demand function and the request price using the following logic: Where (S1,S2,...,S m ) represents m types of services deployed by the device, is the arrival rate of request i; S35. According to the request type vector, the demand function, the request price, and the demand rate of the n resources, the relationship between the price and demand of the single resource is obtained by fitting: F λ =Θ λ (q) S36. Formalizing the resource pricing problem of the device; S37. Determine the optimal benefit of the device using the following logic: S38. Determine the optimal price of the resource using the following logic: q τ,λ =V * (λ,t,n)-V * (λ,t,n-1)+1。 4. The service unloading method based on a multi-level unloading network according to claim 3 is characterized in that: The S36 includes: S361. Use the following logic to express the expected benefit of the device: S362. Using the following logic, the resource pricing problem is transformed into the problem of maximizing the expected benefit of the device: S363, using the following demand function logic, the demand for resource λ is modeled as following a Poisson distribution with an intensity of Φ λ A random process: Where k and α are constants.
5. The service unloading method based on a multi-level unloading network according to claim 1, characterized in that: The task offloading algorithm MOS of the multi-level offloading network in S4 includes: S41, constructing the multi-level offloading network, wherein, from the devices, selecting a suitable auxiliary computing device; identifying a predecessor device and a successor device of the auxiliary computing device to determine a task transfer direction; S42, a user benefit function is formulated to represent the benefit of each intermediate auxiliary computing device, and the price payable by the intermediate auxiliary device to purchase resources from the subsequent device is determined, and the goal of maximizing user benefits is converted into the goal of maximizing the benefit of each device itself, so as to minimize its own cost; S43, executing a multi-level offloading strategy, sorting the pricing of the intermediate auxiliary computing devices, determining and making an offloading task allocation decision based on the resource selling price in the bid table.
6. The service unloading method based on a multi-level unloading network according to claim 5, characterized in that: The S41 includes: S411. Determine the execution time and transmission time of the total task according to the task size C: t e +t i,j ≤τ j In the formula, t e is the total task execution time, t i,j is the transmission time, τ j is the delay constraint of the task received by device j, I and O are the input and output sizes of a single task, r is the transmission rate, and j satisfies the delay constraint; S412, the current device i subtracts t from the delay constraint in the message i,j , and increase the hop count by 1, and send the message to the next device j; S413: When all the successor devices of the current device i cannot meet the delay constraint, determine that the current device i is the last layer node of the multi-level offloading network.
7. The service unloading method based on a multi-level unloading network according to claim 5, characterized in that: The S42 includes: S421, using the following logic, formulate the user benefit function: In the formula, It indicates the value of resources consumed by the user to complete the task alone, and Cost' indicates the value of the resources actually consumed by the user to complete the task. It is the incentive given by the user to device j when the user offloads the task to device j, and V represents the benefit that the user can obtain by completing the task. S422, obtaining the expected rate of return ERR of each intermediate auxiliary computing device. When receiving the bid list of the successor node, the intermediate auxiliary computing device multiplies the bid list by the expected rate of return ERR of the current intermediate auxiliary computing device, and formulates the bid list of the current intermediate auxiliary computing device according to the resource pricing table and the expected rate of return; wherein the income of each intermediate auxiliary computing device is determined by the following logic: W i =R i -Q i The revenue of the auxiliary computing device at the last layer of the multi-level offloading network is used as the resource selling price: W′=Q′ S423: Determine the reward R given by the predecessor node to the device i according to the bid table of the predecessor node according to the device i. i , based on which the following logic is used to determine the price Q payable by the intermediate auxiliary computing device to purchase resources from the successor device i : In the formula, represents the sum of all resource prices of the assigned task completed by device j, q λ,τ represents the price of resource λ in time period τ, o is the number of all resources, Su(i) is the set of successor devices of device i, and Se(j) is the set of time slices for choosing to purchase the resources of device j. S424. Using the following logic, the goal of maximizing user benefits is converted into the goal of maximizing the benefits of each device itself, so as to obtain the minimization of its own cost: minQ i stj∈Su(i) Where C is the size of the total task, is the amount of tasks executed locally on device i, C i is the amount of tasks received by device i.
8. The service unloading method based on a multi-level unloading network according to claim 5, characterized in that: In S43, the price list arrangement operation includes: S431, within the preset time constraint range, according to the type and quantity of resources required by a task in each time slice, calculate the time slice execution task price; S432, according to the time slice with the minimum price for executing the task, select the time slice with the minimum price, calculate the number of tasks m that can be processed by the minimum time slice, fill the price p required by the minimum price time slice into the price table, add the price p to the previous row price, and obtain the price of the min(m-1,n-1) row; remove the current minimum price time slice from the preset time constraint range; S433. Repeat step S432 until the price list is filled with n rows, where n represents the number of tasks.
9. The service unloading method based on a multi-level unloading network according to claim 5, characterized in that: In S43, the operation of calculating the bid table includes: S431', receiving the bid list from all subsequent devices; S432', multiplying the bid table value received by the successor device by the expected rate of return ERR of the intermediate auxiliary calculation device; S433', set |Su(i)|+1 pointer p k , so that p k =1, k∈Su(i). Initialize variables S=0, l=1; S434', according to the value pointed to by the pointer of the current |Su(i)|+1 table row, select the smallest y, and record the table number corresponding to the a table, and perform the update operation using the following logic: S = S + y, T'(l) = S; S435', update the bidding table T a Values: T a (m)=T a (m)-y,m∈{p a +1,p a +2,...,n} Update Table T a Pointer and task number: p = p + 1, l = l + 1; S436', repeat the above steps S434', S435' until l>n.
10. A service offloading system based on a multi-level offloading network, characterized in that: The system comprises: A network model building module, used to build a network model, wherein the devices in the network model include: user equipment and service provider equipment; A network device description module is used to perform abstract operations on the network model, and use preset quadruple to describe the devices in the network model and the task sets generated by the devices. The network device description module is connected to the network model construction module: The pricing algorithm module is used to execute the pricing algorithm of the quantitative resource model, wherein the demand function is obtained and based on the transaction history information, the request price and the demand rate of n resources are calculated to fit the price and demand relationship of a single resource, and the resource pricing problem is converted into the problem of maximizing the expected return of the equipment, so as to obtain the optimal return of the equipment and the optimal price of the resource; The task offloading scheme acquisition module is used to build a multi-level task offloading network with the user device as the center and the maximum task completion delay as the constraint, so as to execute the task offloading algorithm MOS of the multi-level offloading network, perform layer-by-layer quotation operations, and obtain applicable task offloading schemes. The task offloading scheme acquisition module is connected to the pricing algorithm module and the network device description module.
Citation Information
Patent Citations
Demand response service supply method based on reverse auction and D2D communication link
CN114125783A
Mobile user task unloading and resource pricing method and system in edge computing
CN115942384A
Multi-hop multi-layer dynamic task unloading method in Internet of Vehicles
CN115942385A
Low-delay method and system for multi-level edge computing of wireless forward network
CN116033581A
Multi-layer cooperative task unloading optimization method across local edges and cloud resources
CN118175158A
