CN112732442A - Distributed model for edge computing load balancing and solving method thereof - Google Patents
Distributed model for edge computing load balancing and solving method thereof Download PDFInfo
- Publication number
- CN112732442A CN112732442A CN202110031640.9A CN202110031640A CN112732442A CN 112732442 A CN112732442 A CN 112732442A CN 202110031640 A CN202110031640 A CN 202110031640A CN 112732442 A CN112732442 A CN 112732442A
- Authority
- CN
- China
- Prior art keywords
- agent
- task
- tasks
- neighbor
- current
- 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.)
- Granted
Links
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
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5083—Techniques for rebalancing the load in a distributed system
-
- 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)
- Software Systems (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multi Processors (AREA)
Abstract
The invention provides a distributed model for edge computing load balancing and a solving method thereof, wherein the distributed model for edge computing load balancing comprises the following steps: setting an intelligent agent: setting an intelligent object to show an edge server Si∈S,S={S1,S2,...,SmDenotes a set of m edge servers; setting variables: the variables of an agent are composed of one or several tripletsComposition is carried out; setting a value range: the variable value range of an agent consists of three parts, namely a first part and a task TxE is T; the second step,Representing an agentA certain neighbor intelligenceEnergy body, that value range isThird, for the current agent
Description
Technical Field
The invention relates to the field of computers, in particular to a distributed model for edge computing load balancing and a solving method thereof.
Background
With the coming of the world of everything interconnection and the rapid development of artificial intelligence, the number of network terminals and edge devices and the amount of data generated by the network terminals and the edge devices are increased day by day, and the data are processed at the edge of the network. Traditional cloud computing deploys edge servers in the center of a network, so it is difficult to efficiently process data from network edge devices, and therefore the concept of edge computing has been proposed.
The edge calculation is a novel network calculation model for executing calculation at the network edge, and the core idea is that an edge server with certain calculation and storage capacity is deployed at a network base station which is close to terminal equipment (the network base station is close to a network terminal and is the imaging performance of the network edge), and when a terminal task is sent to a cloud end or a result sent by the cloud end is returned to the terminal, the edge server can provide certain calculation and storage resources for the calculation task at the network edge, so that the task response can be better and faster completed. It is worth noting that edge computing is not an alternative to traditional cloud computing, and the two are complementary relationships. Deployment of edge computing has three major benefits over traditional cloud computing: (1) data are processed at the edge of the network, so that the network bandwidth and the power consumption pressure of a cloud computing center are reduced; (2) the task is directly placed on the edge server to be completed without being uploaded to the central edge server, so that the response speed of the task is greatly improved; (3) the task with stronger privacy is directly stored in the edge server instead of the central edge server, so that the safety of user data is protected to a certain extent.
Despite the many advantages of edge computing, there are still some critical issues to be solved, the more critical one being how the edge server implements load balancing. Since the amount of tasks among the edge servers is different, in some time periods, the task load of some edge servers is too high, and the task load of other edge servers is almost zero, so that the time for processing the task by the too-high edge server is too long, and the resource of the too-low edge server is idle.
Disclosure of Invention
The invention provides a distributed model for edge computing load balancing and a solving method thereof, and solves the problem that an edge server in the prior art cannot realize load balancing.
In order to achieve the purpose, the invention adopts the following technical scheme:
the invention firstly provides a distributed model for balancing the edge computing load, which comprises the following steps:
setting an intelligent agent: setting an intelligent object to show an edge server Si∈S,S={S1,S2,...,SmDenotes a set of m edge servers;
setting variables: the variables of an agent are composed of one or several tripletsIn which T isxE.t denotes a task in a set of tasks, T ═ T1,T2,...,TnDenotes a set of n tasks,representing a task TxThe intelligent agent in which the last round is located,representing a task TxThe number of tuples in one agent represents the number of tasks carried by the current agent;
setting a value range: the variable value range of an agent consists of three parts, namely a first part and a task Txe.T (for task T)xIn other words, it may be any task in the task set T); second, for the previous round of task TxThe intelligent agentEach round of allocation can only allocate tasks to neighboring agents, agentsThe value range of (1) is the current agentA set of neighbors ofRepresenting an agentThe value range of a certain neighbor agent isThird, for the current agentIts value range is all agents;
and (4) setting constraints:
local hard constraint: the storage space of the agent should be greater than or equal to the sum of the storage sizes required by all tasks in its local task set, as follows:in the formula (I), the compound is shown in the specification,representing an agentSet of local tasks in round r, TjRepresenting the tasks in the local set of tasks,representing a task TjThe required storage space is largeThe size of the product is small, and the product is small,representing an agentThe size of the storage space of (a);
the local soft constraints include: internal agent constraints and external agent constraints;
internal restraint of the intelligent body: the maximum completion time of all agents is(the agent evaluates the load capacity according to the maximum task completion time of other agents, because the mode of allocating computing resources inside the agent is to turn the task to be processed by the available CPU, the CPU can calculate the maximum time required for completing all tasks in the task queue (assuming that the task in the CPU processing queue is served first), and then the maximum completion time in all available CPUs in the agent is:the method can be used for reflecting the current computing load condition of the edge server, and particularly, an agent with a short maximum task completion time has fewer tasks or stronger computing power, and an agent with a long maximum task completion time has more tasks or weaker computing power. ) Edge serverCalculated time length ofThe calculation formula of (2) is as follows:in the formula (I), the compound is shown in the specification,as an agentThe computing power of (a) is determined,in the formula, numiRepresenting an agentNumber of CPU, tuple<cj,pj,qj>Representing an agentThe name of the jth CPU in (j) is cjJ computing power p of the CPUjAnd a task queue q to be processed of the jth CPUjTask TjPending task queue q belonging to jth CPUj;
External restraint of the agent: the difference gamma of the calculation time length between all the agents (the maximum task completion time length represents the resource load condition of the current agent, so the difference gamma of the maximum task completion time length between the agents represents the load difference between the two agents), and the agentsAnd an agentDifference gamma between the maximum completion time of the taskij,γij∈γ,γijThe calculation formula is as follows:in the formula (I), the compound is shown in the specification,as an agentTask maximization ofThe length of the completion time is long,as an agentThe maximum completion time of the task(s); task transmission delay delta (task T) between two agentsiHas a certain memory size so that during transmission, transmission will also occur), assuming the transmission rate between agentsIf known, task T can be obtained byiIn an agentTo the agentIs delayedThe calculation formula of (2) is as follows:in the formula (I), the compound is shown in the specification,is TiThe size of the memory required for the task.
The invention also provides a distributed model solving method based on the edge computing load balancing distributed model, which comprises the following steps:
s1, initialization: the agent calculates the time needed by the agent to complete the last task in the agent task set according to the current CPU and task distribution information
S2, EC-MGM cycle:
S21、each agentThe current Value information is transmittedSending the information to the neighbor; and collecting Value information sent by neighbors
S22, the agent calculates the sum of the differences of the maximum task completion time between the agent and all the neighbor agents according to the collected information
S23, the agent combines the bandwidth of the neighbor agent to get a new local task allocation plan according to the CPU information and task load condition of the neighbor agentSo thatAnd temporarily storing a new task allocation scheme;
s24, calculating a maximum Gain message Gain (G) by the cost difference between the new task allocation plan and the old task allocation plani);
S25, hard constraint processing:
s251, judgment: if the current storage capacity of the intelligent agent is not enough to store the current task, giving a maximum Gain message Gain (G)i) The value max, max being greater than the maximum Gain message Gain (G) of the neighbor agentj) Executing a new task allocation scheme, allocating the task to the neighbor agent, and then performing step S261; if the current storage capacity of the agent is enough to store the current task, directly performing step S261;
s26, soft constraint processing:
s261, the agent sends a maximum Gain message Gain (G)i) Sending the information to the neighbor agent and collecting the neighbor maximum Gain information Gain (G) sent by the neighbor agentj);
S262, judging: if the local maximum Gain message Gain (G)i) If the value of the current gain message is larger than the maximum gain message of all the neighbors or the value of the current gain message is the maximum value, updating the current task allocation to be a new task allocation scheme, executing the new task allocation scheme, and then executing S263; if not, go to step S263;
and S263, judging whether convergence or termination conditions are met, if so, ending, and otherwise, performing step S21.
Preferably, step S23 includes the steps of:
s231, passing variableJudging which tasks are still local and which tasks are distributed to neighbor intelligent bodies;
and S232, randomly selecting one or more tasks from the tasks which are still local to be distributed to a neighbor intelligent agent.
Preferably, step S1 is also initialized; w is 0; step S263 includes the steps of: s2631, calculating w as w + 1; s2632, judging whether w is less than wmaxIf yes, the process is ended, otherwise, the process proceeds to step S21.
Compared with the prior art, the invention has the following beneficial effects:
(1) compared with the traditional technology, the computing method is good in real-time performance, and the state of each edge server is monitored in real time, so that task allocation is determined according to the processing capacity of each edge server, the quantity of edge devices and servers is increased sharply along with the rapid development of the Internet of things, the task response is very slow due to the fact that the traditional centralized scheduling mode needs to collect global load information, the neighbor server is explored by each edge server by adopting a neighbor exploration mode, and the centralized scheduling mode is avoided, so that the response is fast; (2) the method has good privacy, the traditional centralized task scheduler needs to collect a large amount of edge server information, and once the scheduler is attacked by a network hacker, a large amount of privacy is leaked; and (3) the distributed model realizes that the original task distribution of a central server is changed into the task distribution of a plurality of edge servers, a data framework required by the task distribution in each edge server is built by setting the distributed model with balanced edge computing load, the original centralized mode (collecting all information and making an overall decision) is changed into the distributed coordination mode of the plurality of edge servers, and each intelligent agent only makes a specified behavior, so that the global target with the same trend as the centralized mode is achieved, the phenomenon that the load is overlarge due to centralized processing is avoided, and the response speed is improved.
Additional advantages, objects, and features of the invention will be set forth in part in the description which follows and in part will become apparent to those having ordinary skill in the art upon examination of the following or may be learned from practice of the invention.
Detailed Description
In order to make the technical means, the creation features, the achievement purposes and the functions of the invention clearer and easier to understand, the following specific embodiments further describe the invention:
the invention firstly provides a distributed model for balancing the edge computing load, which comprises the following steps:
setting an intelligent agent: setting an intelligent object to show an edge server Si∈S,SS={S1,S2,...,SmDenotes the set of m edge servers;
setting variables: the variables of an agent are composed of one or several tripletsIn which T isxE.t denotes a task in a set of tasks, T ═ T1,T2,...,TnDenotes a set of n tasks,representing a task TxThe intelligent agent in which the last round is located,representing a task TxThe number of tuples in one agent represents the number of tasks carried by the current agent;
setting a value range: the variable value range of an agent consists of three parts, namely a first part and a task Txe.T (for task T)xIn other words, it may be any task in the task set T); second, for the previous round of task TxThe intelligent agentEach round of allocation can only allocate tasks to neighboring agents, agentsThe value range of (1) is the current agentA set of neighbors ofRepresenting an agentThe value range of a certain neighbor agent isThird, for the current agentIts value range is all agents;
and (4) setting constraints:
local hard constraint: the storage space of the agent should be greater than or equal to the sum of the storage sizes required by all tasks in its local task set, as follows:in the formula (I), the compound is shown in the specification,representing an agentSet of local tasks in round r, TjRepresenting the tasks in the local set of tasks,representing a task TjThe amount of storage space that is required,representing an agentThe size of the storage space of (a);
the local soft constraints include: internal agent constraints and external agent constraints;
internal restraint of the intelligent body: the maximum completion time of all agents is(the agent evaluates the load capacity according to the maximum task completion time of other agents, because the mode of allocating computing resources inside the agent is to turn the task to be processed by the available CPU, the CPU can calculate the maximum time required for completing all tasks in the task queue (assuming that the task in the CPU processing queue is served first), and then the maximum completion time in all available CPUs in the agent is:the method can be used for reflecting the current calculation load condition of the intelligent agent, specifically, the intelligent agent with shorter task maximum completion time has fewer tasks or stronger calculation capacity, and the intelligent agent with longer task maximum completion time has more tasks or more calculation capacityWeak computing power. ) Intelligent agentCalculated time length ofThe calculation formula of (2) is as follows:in the formula (I), the compound is shown in the specification,as an agentThe computing power of (a) is determined,in the formula, numiRepresenting edge serversNumber of CPU, tuple<cj,pj,qj>Representing an agentThe name of the jth CPU in (j) is cjJ computing power p of the CPUjAnd a task queue q to be processed of the jth CPUjTask TjPending task queue q belonging to jth CPUj;
External restraint of the agent: the difference gamma of the calculation time length between all the agents (the maximum task completion time length reflects the resource load condition of the current agent, so the difference gamma of the maximum task completion time length between the agents reflects the load difference between the two agents), and the agentsAnd an agentDifference gamma between the maximum completion time of the taskij,γij∈γ,γijThe calculation formula is as follows:in the formula (I), the compound is shown in the specification,as an agentThe maximum completion time of the task of (1),as an agentThe maximum completion duration of the task of (1); task transmission delay delta (task T) between two agentsiHas a certain memory size so that during transmission, transmission will also occur), assuming the transmission rate between agentsIf known, task T can be obtained byiIn an agentTo the agentIs delayedThe calculation formula of (2) is as follows:in the formula (I), the compound is shown in the specification,is TiThe size of the memory required for the task.
The invention also provides a distributed model solving method based on the edge computing load balancing distributed model, which comprises the following steps:
s1, initialization: the agent calculates the time needed by the agent to complete the last task in the agent task set according to the current CPU and task distribution information
S2, EC-MGM cycle:
s21, each agentThe current Value information is transmittedSending the information to the neighbor; and collecting Value information sent by neighbors
S22, the agent calculates the sum of the differences of the maximum task completion time between the agent and all the neighbor agents according to the collected information
S23, the agent combines the bandwidth of the neighbor agent to get a new local task allocation plan according to the CPU information and task load condition of the neighbor agentSo thatAnd temporarily storing a new task allocation scheme;
s24, calculating a maximum Gain message Gain (G) by the cost difference between the new task allocation plan and the old task allocation plani);
S25, hard constraint processing:
s251, judgment: if the current storage capacity of the intelligent agent is not enough to store the current task, giving a maximum Gain message Gain (G)i) The value max, max being greater than the maximum Gain message Gain (G) of the neighbor agentj) Executing a new task allocation scheme, allocating the task to the neighbor agent, and then performing step S261; if the current storage capacity of the agent is enough to store the current task, directly performing step S261;
s26, soft constraint processing:
s261, the agent sends a maximum Gain message Gain (G)i) Sending the information to the neighbor agent and collecting the neighbor maximum Gain information Gain (G) sent by the neighbor agentj);
S262, judging: if the local maximum Gain message Gain (G)i) If the value of the current gain message is larger than the maximum gain message of all the neighbors or the value of the current gain message is the maximum value, updating the current task allocation to be a new task allocation scheme, executing the new task allocation scheme, and then executing S263; if not, go to step S263;
and S263, judging whether convergence or termination conditions are met, if so, ending, and otherwise, performing step S21.
Preferably, step S23 includes the steps of:
s231, passing variableJudging which tasks are still local and which tasks are distributed to neighbor intelligent bodies;
and S232, randomly selecting one or more tasks from the tasks which are still local to be distributed to a neighbor intelligent agent.
Step S1 is also initialized; w is 0; step S263 includes the steps of: s2631, calculating w as w + 1; s2632, judging whether w is less than wmaxIf yes, the process is ended, otherwise, the process proceeds to step S21.
The following parameter settings can also be set in the model:
a further quintuple of the system model<S,T,R,G,θ>Is formed by S ═ S1,S2,...,SmDenotes a set of m edge servers;
edge server SiThree criteria are: edge server nameComputing power of edge serversStorage capability of edge serverThat is to
Computing power of edge serversThe method comprises the following steps of counting the CPU cores of the edge server, and processing capacity and task queue of the corresponding CPU, namely:in the formula, numiRepresenting edge servers SiNumber of CPU, tuple<c1,p1,q1>Representing edge servers SiThe name of the first CPU in (1) is c1First CPU computing power p1And a queue q of pending tasks for the first CPU1;
T={T1,T2,...,TnDenotes a set of n tasks, task TiThree criteria are: task nameNumber of instructions required for task completionStorage size required for tasksThat is to
R ═ 1, 2.. k } represents the round of task allocation;
representing the bandwidth between connected edge servers in the network topology,representing edge serversTo edge serverNetwork transmission rate between;
θrthe set of local load tasks representing all edge server task assignments for the current round r is represented by:r is formed by the element of R in the formula,representing edge serversA set of local tasks in r rounds;
a task TiThe transmission between the edge servers requires a certain transmission costThe transmission cost of all tasks in the task allocation process is as follows:
value message: such as tuplesWherein S isiThe CPU information and the storage information of the agent i are displayed;it indicates the maximum completion time of the current task of agent i.
Gain message: the maximum difference between the cost of the task allocation scheme newly acquired by the agent and the cost of the old task allocation scheme is also generally referred to as the maximum Gain message Gain (G)i)。
The multi-agent system is one of important research branches of Distributed artificial intelligence and is used for solving Problems in a Distributed environment, and a Distributed Constraint Optimization Problem (DCOPs) is a basic framework for solving the Problems of the multi-agent system. DCOPs are often used as important and useful abstractions of multi-agent collaboration problems, modeling many practical problems in the multi-agent domain. The method emphasizes local interaction through local agents, so that a global optimal result is obtained, the method is an effective technology for coordinating a plurality of agents to solve the distributed problem, the method is one of research hotspots in the field of distributed artificial intelligence at present, and the method is used for modeling edge computing distributed load balance by using DCOPs.
DCOPs consist of a quadruple < A, X, D, F > where A ═ a1,a2,...,amRepresents a set of m agents, one agent being responsible for assigning values to one or more variables; x ═ X1,x2,...xnDenotes n variables, one controlled by only one agent; d ═ D1,D2,...DmIs a set of value ranges, where the variable xiFrom value range DiTaking a middle value; f ═ F1,f2,...fqIs a set of constraint cost functions, each constraint function fi:Is the k variables associated with the functionEach assignment of (a) is combined to a mapping of non-negative costs. One solution for DCOPs is to minimize the sum of all constraint cost functions and contain the combination of the assignments of all variables, i.e.:
it can be known from the above formula that the values of the agents with constraints in the DCOPs are influenced by the constraint functions, and the local constraint functions influence the global target values. Therefore, the DCOPs are suitable for modeling the distributed problems, and the application provides an edge computing load balancing model based on the DCOPs and an algorithm thereof.
Finally, the above embodiments are only used for illustrating the technical solutions of the present invention and not for limiting, although the present invention has been described in detail with reference to the preferred embodiments, it should be understood by those skilled in the art that modifications or equivalent substitutions can be made on the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all of them should be covered in the claims of the present invention.
Claims (4)
1. A distributed model of edge computing load balancing, comprising:
setting an intelligent agent: setting an intelligent object to show an edge server Si∈S,S={S1,S2,...,SmDenotes a set of m edge servers;
setting variables: the variables of an agent are composed of one or several tripletsIn which T isxE.t denotes a task in a set of tasks, T ═ T1,T2,...,TnDenotes a set of n tasks,representing a task TxThe edge server on which the previous round was located,representing a task TxThe number of tuples in one intelligent agent represents the number of tasks borne by the current edge server;
setting a value range: the variable value range of an agent consists of three parts, namely a first part and a task TxE is T; second, for the previous round of task TxThe intelligent agentEach round of assignment can only assign tasks to neighboring agents, agentsThe value range of (1) is the current agentA neighbor set ofRepresenting an agentThe value range of a certain neighbor agent isThird, for the current agentIts value range is all agents;
and (4) setting constraints:
local hard constraint: the storage space of the agent should be greater than or equal to the sum of the storage sizes required by all tasks in its local task set, as follows:in the formula (I), the compound is shown in the specification,representing an agentLocal set of tasks in round r, TjRepresenting the tasks in the local set of tasks,representing a task TjThe amount of storage space that is required,representing an agentThe size of the storage space of (a);
the local soft constraints include: internal agent constraints and external agent constraints;
internal restraint of the intelligent body: the maximum completion time of all agents isEdge serverCalculated time length of The calculation formula of (2) is as follows:in the formula (I), the compound is shown in the specification,as an agentThe computing power of (a) is determined,in the formula, numiRepresenting an agentNumber of CPU, tuple<cj,pj,qj>Representing an agentThe name of the jth CPU in (j) is cjJ computing power p of the CPUjAnd a task queue q to be processed of the jth CPUjTask TjPending task queue q belonging to jth CPUj;
External restraint of the agent: difference gamma of calculated time length between all agents, agentAnd an agentDifference gamma between the maximum completion time of the taskij,γij∈γ,γijThe calculation formula is as follows:in the formula (I), the compound is shown in the specification,as an agentThe maximum completion time of the task of (1),as an agentThe maximum completion time of the task of (1); the task transmission delay delta between two agents is set as the transmission rate between agentsTask T may be obtained byiIn an agentTo the agentIs delayedThe calculation formula of (2) is as follows:in the formula (I), the compound is shown in the specification,is TiThe size of the memory required for the task.
2. A distributed model solving method for the distributed model of edge computing load balancing according to claim 1, comprising the steps of:
s1, initialization: the agent calculates the time needed by the agent to complete the last task in the agent task set according to the current CPU and task distribution information
S2, EC-MGM cycle:
s21, each agentThe current Value information is transmittedSending the information to the neighbor; and collecting Value information sent by neighbors
S22, the agent calculates the sum of the differences of the maximum task completion time between the agent and all the neighbor agents according to the collected information
S23, the agent combines the bandwidth of the neighbor agent to get a new local task allocation plan according to the CPU information and task load condition of the neighbor agentSo thatAnd temporarily storing a new task allocation scheme;
s24, calculating a maximum Gain message Gain (G) by the cost difference between the new task allocation plan and the old task allocation plani);
S25, hard constraint processing:
s251, judgment: if the current storage capacity of the intelligent agent is not enough to store the current task, giving a maximum Gain message Gain (G)i) The value max, max being greater than the maximum Gain message Gain (G) of the neighbor agentj) Executing a new task allocation scheme, allocating the task to the neighbor agent, and then performing step S261; if the current storage capacity of the agent is enough to store the current task, directly performing step S261;
s26, soft constraint processing:
s261, the agent sends a maximum Gain message Gain (G)i) Sending the information to the neighbor agent and collecting the neighbor maximum Gain information Gain (G) sent by the neighbor agentj);
S262, judging: if the local maximum Gain message Gain (G)i) If the value of the current gain message is larger than the maximum gain message of all the neighbors or the value of the current gain message is the maximum value, updating the current task allocation to be a new task allocation scheme, executing the new task allocation scheme, and then executing S263; if not, go to step S263;
and S263, judging whether convergence or termination conditions are met, if so, ending, and otherwise, performing step S21.
3. The distributed model solution method according to claim 1, wherein step S23 includes the following steps:
s231, passing variableJudging which tasks are still local and which tasks are distributed to neighbor agents;
and S232, randomly selecting one or more tasks from the tasks which are still local to be distributed to a neighbor intelligent agent.
4. The distributed model solution method according to claim 1, wherein step S1 is further initialized; w is 0;
step S263 includes the steps of: s2631. Calculating w as w + 1; s2632, judging whether w is less than wmaxIf yes, the process is ended, otherwise, the process proceeds to step S21.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110031640.9A CN112732442B (en) | 2021-01-11 | 2021-01-11 | Distributed model for edge computing load balancing and solving method thereof |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202110031640.9A CN112732442B (en) | 2021-01-11 | 2021-01-11 | Distributed model for edge computing load balancing and solving method thereof |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112732442A true CN112732442A (en) | 2021-04-30 |
CN112732442B CN112732442B (en) | 2023-08-25 |
Family
ID=75590473
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202110031640.9A Active CN112732442B (en) | 2021-01-11 | 2021-01-11 | Distributed model for edge computing load balancing and solving method thereof |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112732442B (en) |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100228819A1 (en) * | 2009-03-05 | 2010-09-09 | Yottaa Inc | System and method for performance acceleration, data protection, disaster recovery and on-demand scaling of computer applications |
CN108141407A (en) * | 2015-10-21 | 2018-06-08 | 英特尔公司 | Mobile edge calculations dynamic accelerates distribution |
US20190138934A1 (en) * | 2018-09-07 | 2019-05-09 | Saurav Prakash | Technologies for distributing gradient descent computation in a heterogeneous multi-access edge computing (mec) networks |
US20190220703A1 (en) * | 2019-03-28 | 2019-07-18 | Intel Corporation | Technologies for distributing iterative computations in heterogeneous computing environments |
CN111538567A (en) * | 2020-04-26 | 2020-08-14 | 国网江苏省电力有限公司信息通信分公司 | Method and equipment for deploying virtual network function chain on edge equipment |
US20200296155A1 (en) * | 2020-03-27 | 2020-09-17 | Intel Corporation | Method, system and product to implement deterministic on-boarding and scheduling of virtualized workloads for edge computing |
CN112132202A (en) * | 2020-09-18 | 2020-12-25 | 嘉兴学院 | Edge computing collaborative member discovery method based on comprehensive trust evaluation |
CN112188442A (en) * | 2020-11-16 | 2021-01-05 | 西南交通大学 | Vehicle networking data-driven task unloading system and method based on mobile edge calculation |
-
2021
- 2021-01-11 CN CN202110031640.9A patent/CN112732442B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20100228819A1 (en) * | 2009-03-05 | 2010-09-09 | Yottaa Inc | System and method for performance acceleration, data protection, disaster recovery and on-demand scaling of computer applications |
CN108141407A (en) * | 2015-10-21 | 2018-06-08 | 英特尔公司 | Mobile edge calculations dynamic accelerates distribution |
US20190138934A1 (en) * | 2018-09-07 | 2019-05-09 | Saurav Prakash | Technologies for distributing gradient descent computation in a heterogeneous multi-access edge computing (mec) networks |
US20190220703A1 (en) * | 2019-03-28 | 2019-07-18 | Intel Corporation | Technologies for distributing iterative computations in heterogeneous computing environments |
US20200296155A1 (en) * | 2020-03-27 | 2020-09-17 | Intel Corporation | Method, system and product to implement deterministic on-boarding and scheduling of virtualized workloads for edge computing |
CN111538567A (en) * | 2020-04-26 | 2020-08-14 | 国网江苏省电力有限公司信息通信分公司 | Method and equipment for deploying virtual network function chain on edge equipment |
CN112132202A (en) * | 2020-09-18 | 2020-12-25 | 嘉兴学院 | Edge computing collaborative member discovery method based on comprehensive trust evaluation |
CN112188442A (en) * | 2020-11-16 | 2021-01-05 | 西南交通大学 | Vehicle networking data-driven task unloading system and method based on mobile edge calculation |
Non-Patent Citations (2)
Title |
---|
王立珩: "分布式约束优化在边缘计算数据缓存问题中的应用研究", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 10 * |
陈建业: "移动边缘计算环境下的服务迁移策略设计与实现", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 11 * |
Also Published As
Publication number | Publication date |
---|---|
CN112732442B (en) | 2023-08-25 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109684075B (en) | Method for unloading computing tasks based on edge computing and cloud computing cooperation | |
CN113950103B (en) | Multi-server complete computing unloading method and system under mobile edge environment | |
CN110493360B (en) | Mobile edge computing unloading method for reducing system energy consumption under multiple servers | |
CN111445111B (en) | Electric power Internet of things task allocation method based on edge cooperation | |
CN107766135B (en) | Task allocation method based on particle swarm optimization and simulated annealing optimization in moving cloud | |
CN109617826B (en) | Storm dynamic load balancing method based on cuckoo search | |
CN109788046B (en) | Multi-strategy edge computing resource scheduling method based on improved bee colony algorithm | |
Liu et al. | Task scheduling in fog enabled Internet of Things for smart cities | |
CN111556516B (en) | Distributed wireless network task cooperative distribution method facing delay and energy efficiency sensitive service | |
CN110233755B (en) | Computing resource and frequency spectrum resource allocation method for fog computing in Internet of things | |
CN109656713B (en) | Container scheduling method based on edge computing framework | |
CN110717300A (en) | Edge calculation task allocation method for real-time online monitoring service of power internet of things | |
CN115629865B (en) | Deep learning inference task scheduling method based on edge calculation | |
CN113961264B (en) | Intelligent unloading algorithm and system for video monitoring cloud edge cooperation | |
CN114265631A (en) | Mobile edge calculation intelligent unloading method and device based on federal meta-learning | |
CN116700993B (en) | Load balancing method, device, equipment and readable storage medium | |
CN113590307A (en) | Edge computing node optimal configuration method and device and cloud computing center | |
CN111953547A (en) | Heterogeneous base station overlapping grouping and resource allocation method and device based on service | |
CN114938372B (en) | Federal learning-based micro-grid group request dynamic migration scheduling method and device | |
Liu et al. | Task scheduling in cloud computing based on improved discrete particle swarm optimization | |
CN112256413A (en) | Scheduling method and device for edge computing task based on Internet of things | |
CN111158893B (en) | Task unloading method, system, equipment and medium applied to fog computing network | |
Hu et al. | Dynamic task offloading in MEC-enabled IoT networks: A hybrid DDPG-D3QN approach | |
CN113329053B (en) | 5G network virtual mapping method and device based on power service characteristics | |
Zhao et al. | Energy-efficient partial offloading in mobile edge computing under a deadline constraint |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |