CN112256415A - Micro-cloud load balancing task scheduling method based on PSO-GA - Google Patents

Micro-cloud load balancing task scheduling method based on PSO-GA Download PDF

Info

Publication number
CN112256415A
CN112256415A CN202011118823.6A CN202011118823A CN112256415A CN 112256415 A CN112256415 A CN 112256415A CN 202011118823 A CN202011118823 A CN 202011118823A CN 112256415 A CN112256415 A CN 112256415A
Authority
CN
China
Prior art keywords
micro
task
cloud
population
individual
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
Application number
CN202011118823.6A
Other languages
Chinese (zh)
Other versions
CN112256415B (en
Inventor
陈星�
姚泽玮
胡俊钦
杨立坚
林潮伟
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.)
Fuzhou University
Original Assignee
Fuzhou University
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 Fuzhou University filed Critical Fuzhou University
Priority to CN202011118823.6A priority Critical patent/CN112256415B/en
Publication of CN112256415A publication Critical patent/CN112256415A/en
Application granted granted Critical
Publication of CN112256415B publication Critical patent/CN112256415B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F9/46—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]
    • G06F9/5083—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)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
  • Data Exchanges In Wide-Area Networks (AREA)

Abstract

The invention relates to a PSO-GA-based micro-cloud load balancing task scheduling method, which comprises the following steps: step S1, initializing a micro cloud set parameter; step S2, calculating the local task response time of all the micro clouds; step S3, sequencing the micro clouds according to the local task response time of the micro clouds, and dividing the micro cloud set into an overload set and an underload set; and step S4, acquiring an optimal micro-cloud task scheduling scheme by adopting an improved PSOGA algorithm based on the constraint conditions. The method and the device can effectively reduce the response time of the migration task and improve the scheduling efficiency of the micro-cloud load task.

Description

Micro-cloud load balancing task scheduling method based on PSO-GA
Technical Field
The invention belongs to the technical field of micro-clouds, and particularly relates to a PSO-GA-based micro-cloud load balancing task scheduling method.
Background
In recent years, with the continuous development of mobile communication technology and internet of things, the number of mobile devices is increasing in a blowout manner. Through various applications on mobile devices, people's lives are more convenient. However, it is difficult to meet the user's requirements for low latency and high reliability due to the limited computing power, battery and data storage capacity of the mobile device itself. The cloud in the mobile cloud computing has abundant computing resources and data storage space, and can efficiently process task requests of application programs on the mobile equipment, so that the problems of limited computing resources and battery energy consumption of the portable mobile equipment are solved. However, migrating the task on the mobile device to the remote cloud may cause long communication delays, which may degrade the QoS experience for the user.
But the advent of micro-cloud technology solved this problem. A micro-cloud is a cluster of computers that are close to a user and connected through a wireless network. In the three MCC frameworks (mobile equipment, micro cloud and remote cloud respectively), the micro cloud is positioned in the middle layer, so that the mobile equipment can access the micro cloud with lower delay through a wireless network, and further real-time interactive response is achieved. Moreover, when the micro cloud is in a load state, the delay tolerant task can be migrated to a remote cloud for execution. The deployment of the micro cloud infrastructure is similar to that of a wireless access point, and many researchers discuss how to deploy micro clouds in a public wireless metropolitan area network, so that the public can enjoy efficient services provided by the micro clouds. Because of the high population density in large cities, mobile users are not on the go to the cloudiness.
However, one of the main problems that the micro-clouds need to deal with is how to distribute task requests for mobile user migration so that the load among the micro-clouds can be balanced. It is common practice to distribute task requests of users to the nearest cloudlets for execution, but when the number of users in an area is too large, the load of the corresponding cloudlets becomes too large. If a large number of user task requests are suddenly received on the micro cloud, the response time of the task is prolonged, and the experience of the user is affected.
Disclosure of Invention
In view of this, the present invention aims to provide a micro cloud load balancing task scheduling method based on a PSO-GA, which can effectively reduce the response time of migration tasks and improve the efficiency of micro cloud load task scheduling.
In order to achieve the purpose, the invention adopts the following technical scheme:
a PSO-GA-based micro cloud load balancing task scheduling method comprises the following steps:
step S1, initializing a micro cloud set parameter;
step S2, calculating local task response time of all micro clouds
Step S3, sequencing the micro clouds according to the local task response time of the micro clouds, and dividing the micro cloud set into an overload set and an underload set;
and step S4, acquiring an optimal micro-cloud task scheduling scheme by adopting an improved PSOGA algorithm based on the constraint conditions.
Further, the micro-cloud set parameters include a service rate, a number of servers, a task arrival rate of each micro-cloud i, and network delay between the micro-clouds.
Further, the step S4 is specifically:
step S41 random initialization PnumIndividuals and checking whether the positions of the generated individuals meet the conditions;
step S42, iterating the population, and obtaining the global extreme value gbest of the population after N iterations
And step S43, obtaining the optimal task scheduling strategy under the corresponding micro cloud set partition according to the global extreme value gbest of the population.
Further, the step S42 is specifically:
i) after the inertia weight omega is calculated, updating the position and the speed of the individual, and checking the calculated position and speed;
ii) selecting a part of individuals from the population to perform mutation operation according to the mutation probability, and checking the positions of the individuals;
iii) calculating the fitness values corresponding to all individuals;
iv) updating the individual extrema of the particles with the global extrema of the population.
Further, the calculating the fitness values corresponding to all individuals specifically includes:
i) by solving for XkUpdating the task arrival rate λ of all the micro-clouds using the following formula;
Figure RE-GDA0002795542610000031
Figure RE-GDA0002795542610000032
ii) computing task latency of overloaded cloudlets
Figure RE-GDA0002795542610000033
And task latency of underloaded cloudlets
Figure RE-GDA0002795542610000034
And network delay Tnet(j)；
iii) taking the maximum task response time in the micro cloud set at this time as the fitness value of the individual.
Further, the updating of the position and the velocity of the particle specifically includes:
the updating of the position and velocity of the individual is achieved using the following formula,
Vk(t+1)＝ωVk(t)+c1r1(pbestk-Xk(t))+c2r2(gbest-Xk(t)) (17)
Xk(t+1)＝Xk(t)+Vk(t+1) (18)
wherein t represents the number of iterations required for the population, c1And c2Is a learning factor of the population, r1And r2Is a random number between (0, 1), ω is the inertial weight; vk(t) is recorded as a memory term, and represents the velocity V of the last iterationk(t) for the current iteration speed Vk(t +1) magnitude and direction; c. C1r1(pbestk-Xk(t)) is recorded as a self-recognition item, is a vector from the current position to the optimal position of the individual, and represents that the movement of the individual comes from self experience; c. C2r2(gbest-Xk(t)) is recorded as a group recognition item and is a vector from the current position to the optimal position in the group;
and examining the individual according to the following formula;
Figure RE-GDA0002795542610000041
Figure RE-GDA0002795542610000042
equation (13) indicates that the task flow redirected to any one micro cloud i in the overloaded micro cloud set to other underloaded micro clouds must be less than or equal to the task arrival rate lambda of the micro cloud ii(ii) a Equation (14) indicates that the total task flow received by any one micro cloud j in the underloaded micro cloud set must be smaller than the total service rate of the server minus the task arrival rate λ of the serverj。
Further, the update of the particle position and velocity adds an inertia weight linear decreasing strategy on the formula (17), as shown in the following formula
ω＝ωmax-t·(ωmax-ωmin)/N，ω∈[ωmin，ωmax] (19) 。
Further, the updating of the individual extremum of each particle and the global extremum of the population specifically includes:
Figure RE-GDA0002795542610000051
Figure RE-GDA0002795542610000052
wherein, equation (20) represents obtaining the self-optimal position of the kth individual since the t-th iteration, and the arg min function is used for returning the self-position X with the minimum fitness value from the 1 st iteration to the t-th iteration; and (3) expressing the optimal position in the population in the t iteration, and acquiring the position X corresponding to the individual with the minimum fitness value in the population by using an arg min function.
Compared with the prior art, the invention has the following beneficial effects:
the method and the device can effectively reduce the response time of the migration task and improve the scheduling efficiency of the micro-cloud load task.
Drawings
FIG. 1 is a flow chart of the method of the present invention;
FIG. 2 is a task response time for 3 algorithms in different configurations in an embodiment of the present invention;
FIG. 3 shows the running times of 3 algorithms in different configurations according to an embodiment of the present invention.
Detailed Description
The invention is further explained below with reference to the drawings and the embodiments.
In this embodiment, it is assumed that a cloudlet provider has established K cloudlets {1, 2., K } in a certain area, and the cloudlets and the wireless access point are both deployed at fixed positions and are connected by a wireless network, so that data transmission can be performed between any two cloudlets. It is assumed that an application program on a mobile device can be dynamically and arbitrarily divided into a plurality of tasks during application running, and the tasks can be migrated to any one of K micro clouds for execution (tasks of a mobile user can be preferentially migrated to the nearby micro clouds).
Modeling the micro cloud by using an M/M/n queuing model, wherein M/M/n respectively represents that the time interval of the task flow migrated by the user successively arrives is in exponential distribution, the task service time on the micro cloud is in exponential distribution, and a single micro cloud has n servers[16]. That is, it means that the micro cloud i has niA server and a service rate of mu per serveriWhere i ∈ {1, 2,..., K }. In addition, each micro cloud i has an important parameter, namely the task arrival rate lambdai. It means a unit timeAverage number of tasks that arrive within. For each micro-cloud i, sampling from a normal distribution N (5, 2) > 0 to determine a micro-cloud service rate μiAnd sampling from the Poisson distribution with the mean value of 3 to determine the number n of micro cloud serversi. Task arrival rate lambda due to micro cloud iiCannot exceed its total service rate mui·niOtherwise, the queuing waiting time of the micro cloud i is infinitely prolonged, so the task arrival rate lambda isiBy distributing 0 < N (15, 6) < mu from normali·ni-0.25 samples for designation. The average waiting time of the task in each micro cloud i consists of two parts, namely task service time and task queuing time. Calculating task latency T using equation (1)iThis time represents the average task latency of the cloudlet i at a task arrival rate of λ.
Figure RE-GDA0002795542610000061
Wherein, the formula (2) is an Erlang C formula, and is used for calculating the probability that a task cannot be immediately processed in the cloudlet and needs to wait according to the number n of servers and the traffic intensity ρ.
Figure RE-GDA0002795542610000071
Since the service request of the user has randomness, the task arrival rate λ on each micro cloud will be different. The task arrival rate λ for some cloudlets may be relatively large, causing the cloudlets to be in an overload state, resulting in an increase in task latency. And the task arrival rate lambda of other micro clouds is smaller, so that most service resources of the micro clouds are in an idle state. Assuming that a cloudlet can redirect a part of its tasks to any other cloudlet, and using f (i, j) to represent the task flow that cloudlet i is redirected to cloudlet j, there are the following restrictions for f (i, j):
Figure RE-GDA0002795542610000072
Figure RE-GDA0002795542610000073
Figure RE-GDA0002795542610000074
wherein equation (3) indicates that any cloudlei to cloudlej task stream f (i, j) is the negative of the cloudlei to cloudlei task stream, and any cloudlei to itself task stream is zero. Equation (4) indicates that the sum of all task flows sent or received by all the micro clouds is equal to zero, i.e., the task flows between the micro clouds are conservative. Formula (5) shows that the sum of the task flows sent by the micro cloud i must be less than or equal to the task arrival rate lambda of the micro cloud ii(here ignoring the task stream received by the micro cloud i).
Since the clouds need to transmit over the wireless network while redirecting a portion of the task stream to other clouds, this will cause a corresponding network delay. Assuming that all migrated packets are the same size, the network delays incurred in transmitting any packet between a pair of cloudiness will be the same. Using C.epsilon.RK×KTo represent a network delay matrix, where ci，jRepresenting the network delay that the task has traversed from micro cloud i to micro cloud j. The network delay c of the micro cloud is specified by sampling in normal distribution of N (0.15, 0.05) to 0.2, wherein N is more than or equal to 0.1. Thus, the network delay required by the redirection task flow f (i, j) < 0 accepted by the micro cloud i is equal to-f (i, j) · ci，j. The total network delay required for each cloudlet to receive the redirection task is calculated using equation (6).
Figure RE-GDA0002795542610000081
Finally, we use d (i) to represent the average task response time for all tasks performed on the cloudlet i, and calculate by equations (1), (6).
Figure RE-GDA0002795542610000082
Wherein the content of the first and second substances,
Figure RE-GDA0002795542610000083
representing the remaining task arrival rate of the cloudlet i. Since a micro cloud may redirect a part of its task stream to other micro clouds or receive a part of the task stream from other micro clouds, its task arrival rate needs to be recalculated when calculating the corresponding task response time of each micro cloud.
Figure RE-GDA0002795542610000084
In this embodiment, the micro cloud load balancing task scheduling problem is defined as follows. K micro clouds are deployed at fixed positions in a city, and each micro cloud i is provided with niA server with a service rate of muiAnd the task arrival rate on the micro cloud i is lambdai(for convenience, we denote the above parameters collectively by the notation CLD). Our goal is to find a set of task flows f ═ { f (i, j) | i, j ∈ {1, 2,. K } } (the task flow must satisfy the requirements of equations (3), (4), (5), through which all the cloudlets are scheduled. This enables the maximum task response time d (i) in the cloudlets to be minimized while all the cloudlets are in a load balanced state. Namely, formula (9):
Figure RE-GDA0002795542610000091
in this embodiment, the micro clouds are sorted according to their local task response time (the local task response time refers to the time taken by all the micro clouds to process only the local task without performing task distribution).
Then using a given cloudlet p, the cloudlet set is divided into an overloaded cloudlet set Vs and an underloaded cloudlet set Vt with its local task response time (equations (10), (11)), and it is ensured that the task flow can only be redirected from the overloaded cloudlet into the underloaded cloudlet.
Using equation (12) as a representation of the solution, where X (i, j) represents i ∈ V from a cloudletsTo the cloudlet j ∈ VtAnd X is a two-dimensional matrix of | Vs | × | Vt | (equation (12)).
Vs＝{i|Ti(λi)＞Tp(λp)}，i∈{1，2，...，K} (10)
Vt＝{j|Tj(λj)≤Tp(λp)}，j∈{1，2，...，K} (11)
Figure RE-GDA0002795542610000092
For the two-dimensional matrix X (i, j), the following constraints apply:
Figure RE-GDA0002795542610000101
Figure RE-GDA0002795542610000102
wherein, the formula (13) indicates that the task flow redirected from any one micro cloud i in the overloaded micro cloud set to other underloaded micro clouds must be less than or equal to the task arrival rate λ thereofi(ii) a Equation (14) indicates that the total task flow received by any one of the micro clouds j in the underloaded micro cloud set must be smaller than the total service rate of the server minus the task arrival rate λ of the serverj。
Referring to fig. 1, in this embodiment, a method for scheduling a micro cloud load balancing task based on a PSO-GA is provided, which includes the following steps:
step S1, initializing micro cloud set parameters, including the service rate, the number of servers, the task arrival rate and the network delay among the micro clouds of each micro cloud i;
step S2, calculating local task response time of all micro clouds
Step S3, sequencing the micro clouds according to the local task response time of the micro clouds, and dividing the micro cloud set into an overload set and an underload set;
and step S4, acquiring an optimal micro-cloud task scheduling scheme by adopting an improved PSOGA algorithm based on the constraint conditions.
The specific process of the adopted PSOGA algorithm is shown as algorithm 1.
Figure RE-GDA0002795542610000103
Figure RE-GDA0002795542610000111
In this embodiment, the step S4 specifically includes:
step S41: random initialization PnumIndividuals and checking whether the positions of the generated individuals meet the conditions;
step S42: iterating the population, and obtaining the global extreme value gbest of the population after N iterations
Step S43: and obtaining an optimal task scheduling strategy under the corresponding micro cloud set division according to the global extreme value gbest of the population.
Preferably, in this embodiment, an initial population of the PSOGA algorithm is constructed.
First, relevant parameters including a population size Pnum, a maximum iteration number N, and the like need to be set.
Next, an initial population is randomly generated
Figure RE-GDA0002795542610000121
In order to perform the iteration.
For individual XkX in (1)k(i) (representing the ith row of the two-dimensional matrix) according to a uniform distribution
Figure RE-GDA0002795542610000122
To choose atAnd (5) filling the machine number.
If randomly generated individual XkViolating equations (13) and (14), the relevant rows or columns of the two-dimensional matrix are randomly selected and reduced until the individual meets the requirements.
Preferably, in this embodiment, the speed of the individual also needs to be initialized. For individual XkVelocity V ink(i) (representing the ith row of a two-dimensional matrix) using uniform distribution
Figure RE-GDA0002795542610000123
Figure RE-GDA0002795542610000124
To make the designation. Here, the evenly distributed value range selects 10% of the task arrival rate of the overloaded cloudlei.
Preferably, in this embodiment, the step S42 specifically includes:
i) after the inertia weight omega is calculated, updating the position and the speed of the individual, and checking the calculated position and speed;
ii) selecting a part of individuals from the population to perform mutation operation according to the mutation probability, and checking the positions of the individuals;
iii) calculating the fitness values corresponding to all individuals;
iv) updating the individual extrema of the particles with the global extrema of the population.
Preferably, in this embodiment, the calculating the fitness values corresponding to all individuals specifically includes:
i) by solving for XkUpdating the task arrival rate λ of all the micro-clouds using the following formula;
Figure RE-GDA0002795542610000131
Figure RE-GDA0002795542610000132
ii) any of the compute overload cloudinessTransaction latency
Figure RE-GDA0002795542610000133
And task latency of underloaded cloudlets
Figure RE-GDA0002795542610000134
And network delay Tnet(j)；
iii) taking the maximum task response time in the micro cloud set at this time as the fitness value of the individual.
In this embodiment, preferably, the updating of the position and speed of the individual is implemented by the following formula, and the individual is checked according to the formulas (13) and (14);
Vk(t+1)＝ωVk(t)+c1r1(pbestk-Xk(t))+c2r2(gbest-Xk(t)) (17)
Xk(t+1)＝Xk(t)+Vk(t+1) (18)
wherein t represents the number of iterations required for the population, c1And c2Is a learning factor of the population, r1And r2Is a random number between (0, 1), ω is the inertial weight; vk(t) is recorded as a memory term, and represents the velocity V of the last iterationk(t) for the current iteration speed Vk(t +1) magnitude and direction; c. C1r1(pbestk-Xk(t)) is recorded as a self-recognition item, is a vector from the current position to the optimal position of the individual, and represents that the movement of the individual comes from self experience; c. C2r2(gbest-Xk(t)) is recorded as a group recognition item and is a vector from the current position to the optimal position in the group;
preferably, the updating of the particle position and the velocity adds an inertia weight linear decreasing strategy on the formula (17), as follows:
ω＝ωmax-t·(ωmax-ωmin)/N，ω∈[ωmin，ωmax] (19)
after the position and the velocity of the individual are calculated, the individual extreme value of each particle and the global extreme value of the population need to be updated. Equation (20) represents obtaining the self-optimum position of the kth individual since the t-th iteration, and the arg min function is used to return the self-position X with the smallest fitness value from 1 st to the t-th iteration. And (3) expressing the optimal position in the population in the t iteration, and acquiring the position X corresponding to the individual with the minimum fitness value in the population by using an arg min function.
Figure RE-GDA0002795542610000141
Figure RE-GDA0002795542610000142
Finally, the mutation operator of the GA algorithm needs to set the mutation probability R of the individuals in the populationmut. When the population performs iterative operation, if the value is [0, 1 ]]The internally generated random number rand _ num is smaller than the variation probability RmutThe current individual is selected to perform mutation operations. The mutation process is as follows:
first, from the set {1, 2., | VsChoose a subset i1，i2，...，ip}，2≤p≤|VsL, |; from the set {1, 2., | VsChoose a subset of j1，j2，...，jq}，2≤q≤|VtL. Wherein the subset { i }1，i2，...，ipCorresponding to the row index of the two-dimensional matrix X, subset j1，j2，...，jqCorresponding to the column indices of the two-dimensional matrix X. Next, the p × q values of the corresponding position in the feasible solution are reinitialized. Finally, whether the mutated individuals meet the requirements of the formulas (13) and (14) is checked.
Example 1:
in this embodiment, it is assumed that the network delay between the micro clouds is proportional to the physical distance between them. The task arrival rate, the service rate and the network delay among the micro clouds of each micro cloud are in normal distribution, the number of micro cloud servers adopts Poisson distribution with the average value of 3, and the specific parameter setting is listed in Table 1.
TABLE 1 Experimental parameters
Figure RE-GDA0002795542610000151
Figure RE-GDA0002795542610000161
And then selecting two algorithms for comparison, namely a random migration algorithm and a greedy algorithm. In the random migration algorithm, m micro clouds with longer local task response time are randomly selected for task migration, and the migrated task amount is also randomly selected. And the greedy algorithm selects the micro cloud with the maximum local task response time each time to migrate the task to the micro cloud with the minimum local task response time until the task response time of all the micro clouds is less than the given constraint. The above algorithms were all run on a server with a CPU of 2.20GHz Intel (R) i5 and 8GiB RAM, and the 3 algorithms were each executed 10 times in each configuration environment, and then the 10 results were averaged.
Table 2 shows the maximum task response time before and after task scheduling for PSOGA, stochastic migration, and greedy algorithms in different configurations. FIG. 2 more intuitively reflects the minimum task response time achievable by 3 algorithms in a 6 configuration environment.
Table 2 maximum task response time before and after scheduling of 3 algorithms under different configurations
Figure RE-GDA0002795542610000162
Figure RE-GDA0002795542610000171
It can be seen that the method of the present invention has a shorter task response time than the other two algorithms. This is because the PSOGA algorithm searches for the optimal solution in the problem domain, and once a better feasible solution is found, all individuals will be continually drawn towards it, and eventually the algorithm will obtain the optimal solution that can be found in the solution space.
The above description is only a preferred embodiment of the present invention, and all equivalent changes and modifications made in accordance with the claims of the present invention should be covered by the present invention.

Claims (8)

1. A PSO-GA-based micro cloud load balancing task scheduling method is characterized by comprising the following steps:
step S1, initializing a micro cloud set parameter;
step S2, calculating local task response time of all micro clouds
Step S3, sequencing the micro clouds according to the local task response time of the micro clouds, and dividing the micro cloud set into an overload set and an underload set;
and step S4, acquiring an optimal micro-cloud task scheduling scheme by adopting an improved PSOGA algorithm based on the constraint conditions.
2. The PSO-GA-based micro-cloud load balancing task scheduling method of claim 1, wherein the micro-cloud set parameters include a service rate, a number of servers, a task arrival rate of each micro-cloud i, and a network delay between micro-clouds.
3. The PSO-GA-based micro-cloud load balancing task scheduling method of claim 1, wherein the step S4 specifically comprises:
step S41 random initialization PnumIndividuals and checking whether the positions of the generated individuals meet the conditions;
step S42, iterating the population, and obtaining the global extreme value gbest of the population after N iterations
And step S43, obtaining the optimal task scheduling strategy under the corresponding micro cloud set partition according to the global extreme value gbest of the population.
4. The PSO-GA-based micro-cloud load balancing task scheduling method of claim 3, wherein the step S42 specifically comprises:
i) after the inertia weight omega is calculated, updating the position and the speed of the individual, and checking the calculated position and speed;
ii) selecting a part of individuals from the population to perform mutation operation according to the mutation probability, and checking the positions of the individuals;
iii) calculating the fitness values corresponding to all individuals;
iv) updating the individual extrema of the particles with the global extrema of the population.
5. The PSO-GA-based micro-cloud load balancing task scheduling method of claim 4, wherein the calculating of the fitness values corresponding to all individuals specifically includes:
i) by solving for XkUpdating the task arrival rate λ of all the micro-clouds using the following formula;
Figure FDA0002731287760000021
Figure FDA0002731287760000022
ii) computing task latency of overloaded cloudlets
Figure FDA0002731287760000023
And task latency of underloaded cloudlets
Figure FDA0002731287760000024
And network delay Tnet(j)；
iii) taking the maximum task response time in the micro cloud set at this time as the fitness value of the individual.
6. The PSO-GA-based micro-cloud load balancing task scheduling method of claim 4, wherein the updating of the particle position and velocity specifically comprises:
the updating of the position and velocity of the individual is achieved using the following formula,
Figure FDA0002731287760000025
Xk(t+1)＝Xk(t)+Vk(t+1) (18)
wherein t represents the number of iterations required for the population, c1And c2Is a learning factor of the population, r1And r2Is a random number between (0, 1), ω is the inertial weight; vk(t) is recorded as a memory term, and represents the velocity V of the last iterationk(t) for the current iteration speed Vk(t +1) magnitude and direction; c. C1r1(pbestk-Xk(t)) is recorded as a self-recognition item, is a vector from the current position to the optimal position of the individual, and represents that the movement of the individual comes from self experience; c. C2r2(gbest-Xk(t)) is recorded as a group recognition item and is a vector from the current position to the optimal position in the group;
and examining the individual according to the following formula;
Figure FDA0002731287760000031
Figure FDA0002731287760000032
equation (13) indicates that the task flow redirected to any one micro cloud i in the overloaded micro cloud set to other underloaded micro clouds must be less than or equal to the task arrival rate lambda of the micro cloud ii(ii) a Equation (14) indicates that the total task flow received by any one micro cloud j in the underloaded micro cloud set must be smaller than the total service rate of the server minus the task arrival rate λ of the serverj。
7. The PSO-GA-based micro-cloud load balancing task scheduling method of claim 6, wherein the updating of particle position and velocity adds an inertial weight linear decreasing strategy on formula (17), as follows
ω＝ωmax-t·(ωmax-ωmin)/N，ω∈[ωmin，ωmax] (19)。
8. The PSO-GA-based micro-cloud load balancing task scheduling method of claim 4, wherein the updating of the individual extremum of each particle and the global extremum of the population specifically is:
Figure FDA0002731287760000041
Figure FDA0002731287760000042
wherein, equation (20) represents obtaining the self-optimal position of the kth individual since the t-th iteration, and the arg min function is used for returning the self-position X with the minimum fitness value from the 1 st iteration to the t-th iteration; and (3) expressing the optimal position in the population in the t iteration, and acquiring the position X corresponding to the individual with the minimum fitness value in the population by using an arg min function.
CN202011118823.6A 2020-10-19 2020-10-19 Micro cloud load balancing task scheduling method based on PSO-GA Active CN112256415B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202011118823.6A CN112256415B (en) 2020-10-19 2020-10-19 Micro cloud load balancing task scheduling method based on PSO-GA

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202011118823.6A CN112256415B (en) 2020-10-19 2020-10-19 Micro cloud load balancing task scheduling method based on PSO-GA

Publications (2)

Publication Number Publication Date
CN112256415A true CN112256415A (en) 2021-01-22
CN112256415B CN112256415B (en) 2023-08-04

Family

ID=74244959

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202011118823.6A Active CN112256415B (en) 2020-10-19 2020-10-19 Micro cloud load balancing task scheduling method based on PSO-GA

Country Status (1)

Country Link
CN (1) CN112256415B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115118728A (en) * 2022-06-21 2022-09-27 福州大学 Ant colony algorithm-based edge load balancing task scheduling method

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107609650A (en) * 2017-09-20 2018-01-19 河南理工大学 Based on the particle genetic algorithm optimization method, apparatus and system that time load is balanced
CN109491761A (en) * 2018-11-07 2019-03-19 中国石油大学(华东) Cloud computing multiple target method for scheduling task based on EDA-GA hybrid algorithm
US20190266534A1 (en) * 2016-07-20 2019-08-29 Worldline Multi-criteria adaptive scheduling method for a market-oriented hybrid cloud infrastructure
CN110809275A (en) * 2019-11-08 2020-02-18 福州大学 Micro cloud node placement method based on wireless metropolitan area network
CN110851272A (en) * 2019-10-30 2020-02-28 内蒙古农业大学 Cloud task scheduling method based on phagocytic particle swarm genetic hybrid algorithm

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190266534A1 (en) * 2016-07-20 2019-08-29 Worldline Multi-criteria adaptive scheduling method for a market-oriented hybrid cloud infrastructure
CN107609650A (en) * 2017-09-20 2018-01-19 河南理工大学 Based on the particle genetic algorithm optimization method, apparatus and system that time load is balanced
CN109491761A (en) * 2018-11-07 2019-03-19 中国石油大学(华东) Cloud computing multiple target method for scheduling task based on EDA-GA hybrid algorithm
CN110851272A (en) * 2019-10-30 2020-02-28 内蒙古农业大学 Cloud task scheduling method based on phagocytic particle swarm genetic hybrid algorithm
CN110809275A (en) * 2019-11-08 2020-02-18 福州大学 Micro cloud node placement method based on wireless metropolitan area network

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115118728A (en) * 2022-06-21 2022-09-27 福州大学 Ant colony algorithm-based edge load balancing task scheduling method
CN115118728B (en) * 2022-06-21 2024-01-19 福州大学 Edge load balancing task scheduling method based on ant colony algorithm

Also Published As

Publication number Publication date
CN112256415B (en) 2023-08-04

Similar Documents

Publication Publication Date Title
CN109857546B (en) Multi-server mobile edge computing unloading method and device based on Lyapunov optimization
CN107273185B (en) Load balancing control method based on virtual machine
CN108170530B (en) Hadoop load balancing task scheduling method based on mixed element heuristic algorithm
WO2016045515A1 (en) Cloud task scheduling algorithm based on user satisfaction
Haris et al. Mantaray modified multi-objective Harris hawk optimization algorithm expedites optimal load balancing in cloud computing
CN112003660B (en) Dimension measurement method of resources in network, calculation force scheduling method and storage medium
CN111163178B (en) Game theory-based service deployment and task unloading method in edge computing
CN107273211B (en) Data processing method based on virtual machine in cloud computing environment
CN109005211B (en) Micro-cloud deployment and user task scheduling method in wireless metropolitan area network environment
CN112888005B (en) MEC-oriented distributed service scheduling method
Pattanaik et al. Performance study of some dynamic load balancing algorithms in cloud computing environment
CN114024970A (en) Power internet of things work load distribution method based on edge calculation
Eisa et al. Enhancing cloud computing scheduling based on queuing models
Wang et al. Dependency-aware dynamic task scheduling in mobile-edge computing
CN112256415B (en) Micro cloud load balancing task scheduling method based on PSO-GA
Jabbari et al. Heterogenous server placement for delay sensitive applications in green mobile edge computing
Addya et al. A hybrid queuing model for virtual machine placement in cloud data center
Xiang et al. Multi-tenant latency optimization in erasure-coded storage with differentiated services
CN110163000A (en) A kind of intelligent data acqusition system and method based on cloud computing system
CN114785692A (en) Virtual power plant aggregation regulation and control communication network flow balancing method and device
Yang et al. Non-cooperative game approach for task offloading in edge clouds
Patel et al. An efficient load balancing mechanism with cost estimation on GridSim
Jabour et al. An optimized approach for efficient-power and low-latency fog environment based on the PSO algorithm
Sarma et al. A survey on load balancing algorithms in cloud computing
Uchechukwu et al. Scalable analytic models for performance efficiency in the cloud

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