CN113900739A - Calculation unloading method and system under many-to-many edge calculation scene - Google Patents

Calculation unloading method and system under many-to-many edge calculation scene Download PDF

Info

Publication number
CN113900739A
CN113900739A CN202111254759.9A CN202111254759A CN113900739A CN 113900739 A CN113900739 A CN 113900739A CN 202111254759 A CN202111254759 A CN 202111254759A CN 113900739 A CN113900739 A CN 113900739A
Authority
CN
China
Prior art keywords
strategy
unloading
edge
energy consumption
many
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111254759.9A
Other languages
Chinese (zh)
Inventor
史彦军
韩超哲
李佳健
沈卫明
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dalian University of Technology
Original Assignee
Dalian University of Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Dalian University of Technology filed Critical Dalian University of Technology
Priority to CN202111254759.9A priority Critical patent/CN113900739A/en
Publication of CN113900739A publication Critical patent/CN113900739A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/44—Arrangements for executing specific programs
    • G06F9/445—Program loading or initiating
    • G06F9/44594—Unloading
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/48—Program initiating; Program switching, e.g. by interrupt
    • G06F9/4806—Task transfer initiation or dispatching
    • G06F9/4843—Task transfer initiation or dispatching by program, e.g. task dispatcher, supervisor, operating system
    • G06F9/4881—Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
    • G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/01—Probabilistic graphical models, e.g. probabilistic networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/04—Constraint-based CAD
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2111/00—Details relating to CAD techniques
    • G06F2111/08—Probabilistic or stochastic CAD
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02D—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN INFORMATION AND COMMUNICATION TECHNOLOGIES [ICT], I.E. INFORMATION AND COMMUNICATION TECHNOLOGIES AIMING AT THE REDUCTION OF THEIR OWN ENERGY USE
    • Y02D10/00—Energy efficient computing, e.g. low power processors, power management or thermal management

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Software Systems (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Computational Mathematics (AREA)
  • Algebra (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computing Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Physics (AREA)
  • Mathematical Analysis (AREA)
  • Probability & Statistics with Applications (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Computer Hardware Design (AREA)
  • Geometry (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a computation unloading method and a system under a many-to-many edge computation scene, which comprises the following steps: acquiring edge calculation scene data of a multi-edge server under a multi-mobile-terminal scene, and constructing an edge calculation scene based on the edge calculation scene data; constructing an unloading model based on the edge calculation scene, wherein the unloading model is modeled based on a Markov decision process; and solving the unloading model by a strategy iteration method to obtain an optimal unloading strategy. The invention can carry out task unloading modeling on heterogeneous tasks generated by terminals with different properties by aiming at minimizing energy consumption, and obtains the optimal strategy of an unloading method.

Description

Calculation unloading method and system under many-to-many edge calculation scene
Technical Field
The invention relates to the technical field of heterogeneous task unloading, in particular to a computing unloading method and a computing unloading system under a many-to-many edge computing scene.
Background
With the rapid development of science and technology, due to the limited resources and computing performance of the intelligent terminal, the intelligent terminal often faces the situation of insufficient computing capability when processing the tasks of computation intensive type and time sensitive type. For this reason, an edge computing mode using an edge server to process analysis data is developed and forms a complement to the conventional cloud computing mode. However, the edge server often has a light weight characteristic, and how to reasonably utilize the limited computing power of the edge server becomes an important problem to be solved urgently in edge computing. The calculation unloading is a key technology of the edge calculation, and can provide calculation resources for the resource-limited equipment to run calculation-intensive application under the condition of limited calculation capacity, so that the calculation speed is increased, and the energy is saved. In other words, computation offloading in edge computing is offloading a computation task of a terminal to an edge cloud, so as to solve the defects of the terminal in the aspects of resource storage, computation performance, energy and the like.
In the prior art, only the requirement of the data size of the heterogeneous task on the terminal computing capacity is considered to be different, but the influence when the priority relationship exists among a plurality of tasks is ignored, for example, some tasks are preempted, that is, during the processing period of other tasks, the task with the preemptive property can preempt other task resources and process preferentially; in addition, many of the existing researches are carried out for "many-to-one" or "one-to-many" scenes, and in addition to the above cases, the existing researches also include "many-to-many" scenes, and the definition of "many-to-many" is as follows: the information transmission is carried out between the plurality of mobile terminals and the plurality of edge servers, each edge server does not have a fixed information transmission corresponding relation with the mobile terminal, and the situation of the task unloading strategy is adjusted in real time according to the actual situation, and the method is not limited to the number of the mobile terminals and the number of the edge servers. However, in the prior art, the technical scheme for researching under the 'many-to-many' unloading scene is less.
Disclosure of Invention
In order to solve the problems of how to model the unloading scene under the 'many-to-many' scene and the like in the prior art, the invention provides a calculation unloading method under the 'many-to-many' scene, which can perform task unloading modeling on heterogeneous tasks generated by terminals with different properties by aiming at minimizing energy consumption.
In order to achieve the technical purpose, the invention provides a computation unloading method under a many-to-many edge computation scene, which comprises the following steps:
acquiring edge calculation scene data of a multi-edge server under a multi-mobile-terminal scene, and constructing an edge calculation scene based on the edge calculation scene data;
constructing an unloading model based on the edge calculation scene, wherein the unloading model is modeled based on a Markov decision process;
and solving the unloading model by a strategy iteration method to obtain an optimal unloading strategy.
Optionally, the edge calculation scenario data includes an edge server set, a vehicle task set, and a remaining calculation capacity.
Optionally, the process of constructing the edge calculation scene includes:
carrying out position constraint on the edge server set and the vehicle set;
performing a remaining computing power constraint on the remaining computing power;
calculating local processing energy consumption and unloading processing energy consumption based on the vehicle task set, and performing minimum energy consumption constraint based on the local processing energy consumption and the unloading processing energy consumption;
and integrating the position constraint result, the calculation capability constraint result and the minimized energy consumption constraint result to obtain an edge calculation scene.
Optionally, the uninstallation model includes a state space, an action space, a reward function, a policy, and a reward;
the state space comprises a plurality of different states, wherein the states comprise positions of all vehicles, residual computing power of all vehicles and residual computing power of all edge servers;
the action space comprises a plurality of different actions, and the actions comprise local processing and edge server unloading processing actions;
the reward function is constructed based on local processing energy consumption and unloading processing energy consumption;
the strategy is the probability of different actions in different states;
the reward is a total reward calculated by a reward function.
Optionally, the process of solving the unloading model by the policy iteration method includes:
constructing a state value function based on the unloading model, wherein the state value function is the mathematical expected sum of the reward function values of all vehicles when different strategies are adopted in different states, and the reward function values are obtained by calculation of the reward function;
initializing the state and the strategy, calculating a state value function based on the initialized state and the strategy to obtain an initial state value, then iteratively updating the state and the strategy, updating the initial state value based on the iteratively updated state and the strategy in the updating process until the updated state value is converged, and obtaining the strategy under the converged state value, namely the optimal unloading strategy.
In order to better achieve the above technical object, the present invention further provides a computation offloading system in a many-to-many edge computation scenario, including:
the first construction module is used for acquiring edge calculation scene data under a multi-edge server multi-mobile-terminal scene, and constructing an edge calculation scene based on the edge calculation scene data;
the second construction module is used for constructing an unloading model based on the edge calculation scene, wherein the unloading model is modeled based on a Markov decision process;
and the solving module is used for solving the unloading model through a strategy iteration method to obtain an optimal unloading strategy.
Optionally, the edge calculation scenario data in the first building module includes an edge server set, a vehicle task set, and a remaining calculation capacity.
Optionally, the first building block includes:
the first constraint module is used for carrying out position constraint on the edge server set and the vehicle set;
the second constraint module is used for carrying out residual computing capacity constraint on the residual computing capacity;
the third constraint module calculates local processing energy consumption and unloading processing energy consumption based on the vehicle task set, and carries out minimum energy consumption constraint based on the local processing energy consumption and the unloading processing energy consumption;
and the edge calculation scene construction module is used for integrating the position constraint result, the calculation capacity constraint result and the minimized energy consumption constraint result to obtain an edge calculation scene.
Optionally, the uninstallation model in the second building module includes a state space, an action space, a reward function, a policy, and a reward;
the state space comprises a plurality of different states, wherein the states comprise positions of all vehicles, residual computing power of all vehicles and residual computing power of all edge servers;
the action space comprises a plurality of different actions, and the actions comprise local processing and edge server unloading processing actions;
the reward function is constructed based on local processing energy consumption and unloading processing energy consumption;
the strategy is the probability of different actions in different states;
the reward is a total reward calculated by a reward function.
Optionally, solving the module packet block:
the first processing module is used for constructing a state value function based on the unloading model, wherein the state value function is the mathematical expected sum of the reward function values of all vehicles when different strategies are adopted in different states, and the reward function values are obtained by calculation of the reward function;
the second processing module is used for initializing the state and the strategy, calculating a state value function based on the initialized state and the strategy, acquiring an initial state value, then performing iterative update on the state and the strategy, updating the initial state value based on the iteratively updated state and strategy in the updating process until the updated state value is converged, and acquiring the strategy under the converged state value, namely the optimal unloading strategy.
The invention has the following technical effects:
the method and the system provided by the invention consider the size of the data volume of the task and the priority relation between the tasks, take the minimization of energy consumption as a target, carry out the unloading of the heterogeneous tasks under the scene of many-to-many, can accurately establish the related optimal unloading strategy, can minimize the energy consumption in the task unloading process, and have strong practicability.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
Fig. 1 is a flowchart of a problem solving idea used for solving an optimal allocation policy according to a first embodiment of the present invention;
fig. 2 is a schematic diagram illustrating distribution of a lower edge server and terminals in a scenario where multiple terminals correspond to a multi-edge server according to an embodiment of the present invention;
fig. 3 is a schematic diagram of a problem solving system used for solving the optimal allocation strategy according to the second embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
In order to solve the problem of how to model an unloading scene under a 'many-to-many' scene in the prior art, the invention provides the following scheme:
example one
As shown in fig. 1-2, the present invention provides a method for offloading computing tasks in an edge computing scenario with multiple mobile terminals and multiple edge servers. In the running process of a plurality of terminals, aiming at the generated heterogeneous tasks, unloading modeling is carried out from the two aspects of task priority and the size of the tasks by taking terminal energy consumption minimization as a target, and finally, a strategy iteration method is utilized and a greedy algorithm is adopted to solve the established unloading model so as to obtain the optimal unloading strategy in each state.
Since the present invention is mainly directed to the unloading of heterogeneous tasks generated by various AGV vehicles in an industrial park, the mobile terminals referred to below are all represented by various AGV vehicles. The specific contents are as follows:
step 1: the method comprises the steps of obtaining edge calculation scene data under a multi-edge server multi-mobile-terminal scene, and constructing an edge calculation scene, namely constructing the edge calculation scene of the multi-edge server multi-mobile-terminal based on the edge calculation scene data
In an industrial park covered by a plurality of edge servers, a plurality of AGV vehicles perform transportation tasks on the park. Assume that a total of m edge servers form a set EC ═ EC1,EC2,EC3KECmThe components are uniformly distributed in an industrial park; the number of AGV vehicles is n, and one set of AGV is formed as the { AGV1,AGV2,AGV3KAGVn}；AGViMultiple tasks may be generated at a time, the AGViThe generated task set is Ti＝{Ti1,Ti2,Ti3K }, and T }i∈T＝{T1,T2KTnRepresents the total set of tasks generated by all AGV vehicles.
After all tasks are generated, two treatment modes exist: local processingAnd offloaded to some edge server for processing. Task
Figure BDA0003323726220000071
Wherein ω represents the urgency of the task dividing the task into: urgent tasks and non-urgent tasks, respectively, are labeled {0, 1} by the AGV vehicles. For the emergency task, considering that the delay time requirement of the task is high, the emergency task is directly processed locally, and the computing resources of the non-emergency task can be preempted; and for the non-emergency tasks, a Markov decision process modeling method is utilized to make a decision with the aim of minimizing the energy consumption of the AGV, so as to judge the processing mode of each task.
Figure BDA0003323726220000072
Indicating the size of the data volume of the task. Theta represents a priority value after the task quantization, the priority value is determined according to the emergency degree of the task, the sensitivity degree of the task to delay and the influence on the overall operation of the industrial park, and the task with high priority is preferentially processed in a task cache queue in the scheduling process.
Any AGV vehicle runs according to the AGViLocation and remaining computing power of
Figure BDA0003323726220000081
ECjThe remaining computing power and the amount of energy consumed to offload to each edge server determine which edge server needs to offload tasks to, i.e., the tasks generated by the AGV vehicles can be offloaded to any one of the edge servers nearby as needed, but need to satisfy:
Figure BDA0003323726220000082
l represents the distance between the vehicle and the edge server to which the task is ultimately offloaded; l0Signal maximum coverage radius representing all edge servers;
Figure BDA0003323726220000083
representing the remaining computing power of each edge server;
Figure BDA0003323726220000084
indicating the remaining computing power of each AGV vehicle.
Because the patent is oriented to the unloading of the calculation tasks of the AGV vehicles in the industrial park, the AGV vehicles need to be charged every time when working for a period of time, and in order to fully utilize the energy of the battery, the energy of the AGV vehicles needs to be preferentially ensured in the calculation process, namely the energy consumption of the AGV vehicles is minimum; the edge server has a continuous energy source, so the energy consumption of the edge server is not considered in the calculation. Regarding energy consumption in the transmission process, no energy consumption in the transmission process exists in the local processing process, and when the calculation result is returned to the AGV after being unloaded to the edge server, the energy consumption can be basically ignored due to the fact that the data volume of the calculation result is small. In summary, only the energy consumption calculated locally by the AGV vehicles and the energy consumption of the unloading process to the edge server are considered herein.
Namely energy consumption under the condition of local treatment
Figure BDA0003323726220000085
CperRepresenting the number of CPU cycles, P, required for the AGV vehicle to compute locally per unit dataperRepresenting energy consumption per cycle of AGV vehicle local calculation
② energy consumption under condition of task unloading to edge server
Figure BDA0003323726220000091
PcsRepresenting transmission power, v representing data transmission rate
Step 2: based on the edge calculation scene, an unloading model is constructed, namely a task unloading model of 'many-to-many' edge calculation is established
The task unloading modeling method commonly used in the edge calculation at present comprises a convex optimization method, a game theory method, a Markov decision method, a Lyapunov optimization method and other methods, and considering that the scene is mainly used for unloading calculation tasks of various AGV vehicles in an industrial park, new tasks can be continuously generated in the scene, the state is continuously changed, the state at the next moment is irrelevant to the historical state and is only relevant to the state at the previous moment, in the step 2, the Markov decision process is adopted for modeling, the quality of each strategy is converted into the data size of final return, and the optimal solution scheme is given under the condition of sufficient calculation resources; otherwise, an approximate solution of the optimal solution which can be accepted by people can be provided.
The unloading model mainly comprises five tuples { S, A, R, pi, R }, wherein S is a state space, and a state S is in the time periodt＝{x,y,fAGV,fECThe position of each AGV vehicle, the residual computing capacity of each AGV vehicle and the residual computing capacity of each edge server are included; a is an action space, which is composed of actions of all vehicles in the time period, and a ═ a1,a2,a3...anAnd A represents a set of processing modes of all tasks generated by each AGV vehicle, and the processing modes comprise: local processing and offloading to some edge server processing; r is a reward function, which is mainly a function related to the total energy consumption of each AGV vehicle processing task; pi is a strategy and represents the probability of the AGV vehicle making various actions under various states; r is the return, representing the total prize ultimately won. For the reward function R, the following is defined:
for local processing task (two cases according to whether the upper limit of the calculation capacity of the AGV vehicle is exceeded)
Figure BDA0003323726220000101
② the task of unloading to the edge cloud server (the upper limit of the computation capability of the unloaded edge cloud server is divided into two cases according to whether the upper limit is exceeded)
Figure BDA0003323726220000102
Since the final goal of using the markov decision process in the present invention is: determining whether all tasks generated by each AGV vehicle are offloaded to an edge server for processing or processed locally, so that a final return equal to the sum of the reward functions of each AGV vehicle is obtained:
Figure BDA0003323726220000103
in the process of unloading the tasks, the residual computing power of each AGV needs to be considered, and the residual computing power of each edge server, that is, the tasks received by each edge server or the tasks processed by the AGV by itself cannot exceed the upper limit of the processing power.
And step 3: solving the unloading model by a strategy iteration method to obtain an optimal unloading strategy, namely determining the optimal unloading strategy
And solving an optimal strategy by adopting a strategy iteration method for the obtained unloading model. Each state in the Markov decision process corresponds to a state value function:
Vπ(si)＝Eπ[∑Ri|si]
the function is expressed by when the state is siAnd when the strategy adopted is pi, the sum of the mathematical expectations of the reward function values R after all tasks generated by each AGV vehicle make a decision.
The method then gives an initial state s0And initial strategy pi0Calculating s under the strategy0And performing strategy improvement according to the V(s) value, and then continuing to perform a new strategy improvement according to a new V(s) value until the strategy is converged finally, so that an approximate solution of the optimal strategy can be obtained efficiently by using a greedy algorithm.
In a random initial state s0And initial strategy pi0The following:
step 3-1: randomly selecting an AGV, only adjusting a task allocation strategy of the AGV, simultaneously ensuring the upper limit of the computing capacity of each server and the upper limit of the computing capacity of the AGV, and enabling V(s) to reach the maximum value on the premise of not changing the allocation strategy of other AGV vehicles;
step 3-2: just like the step 3-1, the task allocation strategy of only one AGV vehicle is randomly changed, but the allocation strategy of the AGV vehicles in the step one is kept unchanged;
step 3-3: just changing the task allocation strategy of one AGV at random as in the step 3-2, but keeping the allocation strategy of the AGV selected in the step one and the step two unchanged;
by analogy, after one cycle, a better than initial strategy pi is finally obtained0A policy of (1).
The next cycle is started: and (3) changing the AGV vehicle selected in the step one, obtaining a better strategy under various conditions, comparing and selecting the strategy with the maximum V(s) value, and at the moment, minimizing the total energy consumption of the terminal under the strategy.
For ease of understanding of the description, the present invention provides the following examples: in an industrial park covered by a plurality of edge servers, a plurality of AGV vehicles perform transportation tasks on the park. Assume that there are 3 edge servers in total forming a set EC ═ EC1,EC2,EC3The components are uniformly distributed in an industrial park; the number of AGV vehicles is 5, and one set of AGV is formed as the AGV1,AGV2,AGV3,AGV4,AGV5}；AGViMultiple tasks may be generated at a time, the AGViThe generated task set is Ti＝{Ti1,Ti2,Ti3K }, and T }i∈T＝{T1,T2KTnRepresents the total set of tasks generated by all AGV vehicles.
According to the scene, an unloading model is established by utilizing a Markov decision process, and the unloading model mainly comprises five tuples { S, A, R, pi, R }, wherein S is a state space, A is an action space, R is a reward function, and pi is a strategy R and is a return.
After the task is generated, the state is s0Random generation strategy pi0：
Step 3-1: randomly selecting an AGV (e.g., AGV)1) Only adjust AGV1Sub-strategy for a single vehicle, while requiring each server to be guaranteedThe tasks distributed by the AGV and the vehicle can not reach the upper limit of the computing capacity of each server and the upper limit of the computing capacity of the AGV vehicle, and all other AGV vehicles (AGV vehicles) are not changed2,AGV3,AGV4,AGV5) Maximizing v(s) on the premise of a sub-allocation policy;
step 3-2: just like step 3-1, only one AGV vehicle (e.g., AGV) is randomly changed2) The sub-strategy of (1) to maximize V(s), but step one AGV1The sub-policy of (A) is kept consistent with the result of step one and the AGV3,AGV4,AGV5The sub-policies of (a) remain unchanged;
step 3-3: just like step 3-2, only one AGV vehicle (e.g., AGV) is randomly changed3) The task allocation strategy of (1) maximizes V(s), but the AGV selected in step one and step two1,AGV2The sub-strategy of (2) is kept consistent with the step (II);
by analogy, after one cycle, a better than initial strategy pi is finally obtained0A policy of (1).
The next cycle is started: changing the AGV vehicle selected in step one (e.g., the initially selected vehicle is an AGV)2,AGV3,AGV4,AGV5Any of these) can get better strategy under each situation, compare the results, select the strategy with the largest v(s) value, at which time the total energy consumption of the terminal is the smallest.
Example two
To better achieve the above technical object, as shown in fig. 3, the present invention further provides a computation offloading system in a many-to-many edge computation scenario, including:
the first construction module is used for acquiring edge calculation scene data under a multi-edge server multi-mobile-terminal scene, and constructing an edge calculation scene based on the edge calculation scene data;
the second construction module is used for constructing an unloading model based on the edge calculation scene, wherein the unloading model is modeled based on a Markov decision process;
and the solving module is used for solving the unloading model through a strategy iteration method to obtain an optimal unloading strategy.
Optionally, the edge calculation scenario data in the first building module includes an edge server set, a vehicle task set, and a remaining calculation capacity.
Optionally, the first building block includes:
the first constraint module is used for carrying out position constraint on the edge server set and the vehicle set;
the second constraint module is used for carrying out residual computing capacity constraint on the residual computing capacity;
the third constraint module calculates local processing energy consumption and unloading processing energy consumption based on the vehicle task set, and carries out minimum energy consumption constraint based on the local processing energy consumption and the unloading processing energy consumption;
and the edge calculation scene construction module is used for integrating the position constraint result, the calculation capacity constraint result and the minimized energy consumption constraint result to obtain an edge calculation scene.
Optionally, the uninstallation model in the second building module includes a state space, an action space, a reward function, a policy, and a reward;
the state space comprises a plurality of different states, wherein the states comprise positions of all vehicles, residual computing power of all vehicles and residual computing power of all edge servers;
the action space comprises a plurality of different actions, and the actions comprise local processing and edge server unloading processing actions;
the reward function is constructed based on local processing energy consumption and unloading processing energy consumption;
the strategy is the probability of different actions in different states;
the reward is a total reward calculated by a reward function.
Optionally, solving the module packet block:
the first processing module is used for constructing a state value function based on the unloading model, wherein the state value function is the mathematical expected sum of the reward function values of all vehicles when different strategies are adopted in different states, and the reward function values are obtained by calculation of the reward function;
the second processing module is used for initializing the state and the strategy, calculating a state value function based on the initialized state and the strategy, acquiring an initial state value, then performing iterative update on the state and the strategy, updating the initial state value based on the iteratively updated state and strategy in the updating process until the updated state value is converged, and acquiring the strategy under the converged state value, namely the optimal unloading strategy. In addition, the system and the method of the present invention correspond to each other, and the detailed explanation of the method will not be described herein.
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (10)

1. A computation offloading method under a many-to-many edge computation scenario is characterized by comprising:
acquiring edge calculation scene data of a multi-edge server under a multi-mobile-terminal scene, and constructing an edge calculation scene based on the edge calculation scene data;
constructing an unloading model based on the edge calculation scene, wherein the unloading model is modeled based on a Markov decision process;
and solving the unloading model by a strategy iteration method to obtain an optimal unloading strategy.
2. The method of claim 1, wherein the method comprises:
the edge calculation scene data comprises an edge server set, a vehicle task set and residual calculation capacity, wherein vehicle tasks in the vehicle task set comprise whether the vehicle tasks are emergency tasks or not, data size and task priority.
3. The method of claim 2, wherein the method comprises:
the process of constructing the edge calculation scene comprises the following steps:
carrying out position constraint on the edge server set and the vehicle set;
performing a remaining computing power constraint on the remaining computing power;
calculating local processing energy consumption and unloading processing energy consumption based on the vehicle task set, and performing minimum energy consumption constraint based on the local processing energy consumption and the unloading processing energy consumption;
and integrating the position constraint result, the calculation capability constraint result and the minimized energy consumption constraint result to obtain an edge calculation scene.
4. The method of claim 1, wherein the method comprises:
the unloading model comprises a state space, an action space, a reward function, a strategy and a return;
the state space comprises a plurality of different states, wherein the states comprise positions of all vehicles, residual computing power of all vehicles and residual computing power of all edge servers;
the action space comprises a plurality of different actions, and the actions comprise local processing and edge server unloading processing actions;
the reward function is constructed based on local processing energy consumption and unloading processing energy consumption;
the strategy is the probability of different actions in different states;
the reward is a total reward calculated by a reward function.
5. The computation offload method under the many-to-many edge computation scenario of claim 4, wherein:
the process of solving the unloading model by the strategy iteration method comprises the following steps:
constructing a state value function based on the unloading model, wherein the state value function is the mathematical expected sum of the reward function values of all vehicles when different strategies are adopted in different states, and the reward function values are obtained by calculation of the reward function;
initializing the state and the strategy, calculating a state value function based on the initialized state and the strategy to obtain an initial state value, then iteratively updating the state and the strategy, updating the initial state value based on the iteratively updated state and the strategy in the updating process until the updated state value is converged, and obtaining the strategy under the converged state value, namely the optimal unloading strategy.
6. The system for computation offloading of a computation offloading method under a many-to-many edge computation scenario of any of claims 1-5, comprising:
the first construction module is used for acquiring edge calculation scene data under a multi-edge server multi-mobile-terminal scene, and constructing an edge calculation scene based on the edge calculation scene data;
the second construction module is used for constructing an unloading model based on the edge calculation scene, wherein the unloading model is modeled based on a Markov decision process;
and the solving module is used for solving the unloading model through a strategy iteration method to obtain an optimal unloading strategy.
7. The computing offload system under many-to-many edge computing scenario of claim 6, wherein:
the edge calculation scene data in the first construction module comprises an edge server set, a vehicle task set and residual calculation capacity.
8. The computing offload system under many-to-many edge computing scenario of claim 7, wherein:
the first building block comprises:
the first constraint module is used for carrying out position constraint on the edge server set and the vehicle set;
the second constraint module is used for carrying out residual computing capacity constraint on the residual computing capacity;
the third constraint module calculates local processing energy consumption and unloading processing energy consumption based on the vehicle task set, and carries out minimum energy consumption constraint based on the local processing energy consumption and the unloading processing energy consumption;
and the edge calculation scene construction module is used for integrating the position constraint result, the calculation capacity constraint result and the minimized energy consumption constraint result to obtain an edge calculation scene.
9. The computing offload system under many-to-many edge computing scenario of claim 6, wherein:
the unloading model in the second construction module comprises a state space, an action space, a reward function, a strategy and a return;
the state space comprises a plurality of different states, wherein the states comprise positions of all vehicles, residual computing power of all vehicles and residual computing power of all edge servers;
the action space comprises a plurality of different actions, and the actions comprise local processing and edge server unloading processing actions;
the reward function is constructed based on local processing energy consumption and unloading processing energy consumption;
the strategy is the probability of different actions in different states;
the reward is a total reward calculated by a reward function.
10. The computing offload system under many-to-many edge computing scenario of claim 9, wherein:
solving module packet blocks:
the first processing module is used for constructing a state value function based on the unloading model, wherein the state value function is the mathematical expected sum of the reward function values of all vehicles when different strategies are adopted in different states, and the reward function values are obtained by calculation of the reward function;
the second processing module is used for initializing the state and the strategy, calculating a state value function based on the initialized state and the strategy, acquiring an initial state value, then performing iterative update on the state and the strategy, updating the initial state value based on the iteratively updated state and strategy in the updating process until the updated state value is converged, and acquiring the strategy under the converged state value, namely the optimal unloading strategy.
CN202111254759.9A 2021-10-27 2021-10-27 Calculation unloading method and system under many-to-many edge calculation scene Pending CN113900739A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111254759.9A CN113900739A (en) 2021-10-27 2021-10-27 Calculation unloading method and system under many-to-many edge calculation scene

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111254759.9A CN113900739A (en) 2021-10-27 2021-10-27 Calculation unloading method and system under many-to-many edge calculation scene

Publications (1)

Publication Number Publication Date
CN113900739A true CN113900739A (en) 2022-01-07

Family

ID=79027163

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111254759.9A Pending CN113900739A (en) 2021-10-27 2021-10-27 Calculation unloading method and system under many-to-many edge calculation scene

Country Status (1)

Country Link
CN (1) CN113900739A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115016932A (en) * 2022-05-13 2022-09-06 电子科技大学 Embedded distributed deep learning model resource elastic scheduling method

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115016932A (en) * 2022-05-13 2022-09-06 电子科技大学 Embedded distributed deep learning model resource elastic scheduling method

Similar Documents

Publication Publication Date Title
CN112188442B (en) Vehicle networking data-driven task unloading system and method based on mobile edge calculation
CN112004239A (en) Computing unloading method and system based on cloud edge cooperation
CN112911016B (en) Edge-side collaborative computing unloading method and system, electronic equipment and storage medium
CN112104502A (en) Time-sensitive multitask edge computing and cache cooperation unloading strategy method
CN112153145A (en) Method and device for unloading calculation tasks facing Internet of vehicles in 5G edge environment
CN115297171B (en) Edge computing and unloading method and system for hierarchical decision of cellular Internet of vehicles
CN113316116B (en) Vehicle calculation task unloading method
CN110570075A (en) Power business edge calculation task allocation method and device
CN112612553A (en) Container technology-based edge computing task unloading method
CN113641417A (en) Vehicle safety task unloading method based on branch-and-bound method
CN113900739A (en) Calculation unloading method and system under many-to-many edge calculation scene
CN113709249B (en) Safe balanced unloading method and system for driving assisting service
CN113315669B (en) Cloud edge cooperation-based throughput optimization machine learning inference task deployment method
CN114626298A (en) State updating method for efficient caching and task unloading in unmanned aerial vehicle-assisted Internet of vehicles
Safavifar et al. Adaptive workload orchestration in pure edge computing: A reinforcement-learning model
CN116781788B (en) Service decision method and service decision device
CN112398917A (en) Real-time task scheduling method and device for multi-station fusion architecture
CN112559078A (en) Method and system for hierarchically unloading tasks of mobile edge computing server
CN111930435A (en) Task unloading decision method based on PD-BPSO technology
CN115988462A (en) Debugging method of edge computing module based on vehicle-road cooperation
CN116996941A (en) Calculation force unloading method, device and system based on cooperation of cloud edge ends of distribution network
CN116112981A (en) Unmanned aerial vehicle task unloading method based on edge calculation
CN115665869A (en) Multi-user collaboration platform and method based on edge calculation and directed acyclic graph
Li et al. A cost‐efficient and QoS‐aware adaptive placement of applications in fog computing
CN114741191A (en) Multi-resource allocation method for compute-intensive task relevance

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination