CN114338662A - Task unloading and resource allocation method based on user fairness maximization - Google Patents
Task unloading and resource allocation method based on user fairness maximization Download PDFInfo
- Publication number
- CN114338662A CN114338662A CN202111325086.1A CN202111325086A CN114338662A CN 114338662 A CN114338662 A CN 114338662A CN 202111325086 A CN202111325086 A CN 202111325086A CN 114338662 A CN114338662 A CN 114338662A
- Authority
- CN
- China
- Prior art keywords
- task
- unloading
- food source
- resource allocation
- scheme
- 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
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a task unloading and resource allocation method based on user fairness maximization.A marginal computing service platform collects task information of a corresponding region, takes the maximized user fairness as an objective function of the marginal computing service platform, unloads tasks in a task set to a marginal computing server for computation, and correspondingly allocates computing resources required by the tasks. Specifically, the edge service platform collects task information of the time slot at the beginning of each time slot, firstly selects an edge calculation server for each task, and determines the quantity of resources obtained by each task on a corresponding server by adopting a resource allocation method with maximized fairness on the basis of a task unloading scheme so as to minimize the maximum service experience deficiency coefficient of all tasks in the time slot. The method of the invention can make the user have better service experience under the condition of budget constraint.
Description
Technical Field
The invention relates to the technical field of edge computing, in particular to a task unloading and resource allocation method based on user fairness maximization.
Background
Task offloading in edge computing is a technique for offloading all or part of a compute-intensive task to a cloud server with sufficient resources by a resource-constrained mobile device, and mainly plays a role in saving energy and battery life or an application program which cannot be processed heavily by a terminal device.
In the existing research, the task unloading optimization of a three-layer framework of user equipment, an edge computing server and a cloud server is mostly discussed, although the cloud server has massive computing resources, the cloud server is located at a remote geographical position, and transmission delay which cannot be accepted by a user is caused in the unloading process. In the existing work, the task unloading and resource allocation scheme in the edge calculation designed by researchers mainly considers optimization time delay, energy, resource utilization rate and the like, and neglects the importance of user fairness. Therefore, there is a need to provide a task offloading and resource allocation scheme that does not rely on cloud-side servers, but solves this problem by making reasonable use of resources in edge server farms that are geographically relatively close, while taking user fairness into account.
Disclosure of Invention
The invention aims to overcome the defects and shortcomings of the prior art, and provides a task unloading and resource allocation method based on user fairness maximization, which can guarantee the fair allocation of resources among different tasks under the condition of budget constraint and guarantee the high utilization rate of computing resources of an edge computing system.
In order to achieve the purpose, the technical scheme provided by the invention is as follows: a task unloading and resource allocation method based on user fairness maximization considers that one server comprises m edge calculation servers
Region of (1), MmFor the mth edge, a server is calculated in which n tasks are distributed in the region
TnFor the nth task, each task needs to be unloaded to the edge computing server and computed by using computing resources provided by the edge computing server; t isiFor the ith task, task TiService experience deficiency coefficient GiDefined as the actual time delay diWith maximum allowable delay DiRatio of (i.e. G)i＝di/DiWherein task TiActual time delay d ofiThe task set is the sum of the transmission and propagation delay of the task and the calculation delay on the edge calculation server
The maximum service experience deficiency factor in is
The method is to minimize task set
Maximum service experience deficiency factor in
For the purpose of determining the unloading place and the quantity of the distributed computing resources for each task, the target function of the edge computing service platform is designed as follows:
wherein gamma is the unloading scheme of the task,
resource allocation schemes for tasks, FcompComputing a server M for an edgecompTotal amount of computing resources of fcomp,iIs McompAssigned to task TiNumber of computing resources of SiFor task TiThe type of service that is required is,
is McompA set of offered service types; the limiting conditions are respectively as follows: mcompThe sum of the computing resources allocated to the task does not exceed McompTotal number of computing resources, task TiActual time delay d ofiNot exceeding TiMaximum allowable delay DiAnd the unloading place of the task needs to select the edge computing server configured with the required service type;
the edge computing service platform is a general name of all edge computing servers, namely the edge computing service platform is formed by all the edge computing servers.
The task unloading and resource allocation method based on the user fairness maximization comprises the following steps:
1) initializing a population with NP food sources, wherein each food source corresponds to a task unloading scheme and a corresponding resource allocation scheme; then, a resource allocation method for guaranteeing user fairness, called FGRA, is used, and the resource allocation method calculates a corresponding resource allocation scheme and a corresponding maximum service experience deficiency coefficient for each task unloading scheme
2) Performing Cycle round optimization on food sources in a population by using an evolutionary algorithm combining the ideas of a swarm algorithm, a genetic algorithm and a particle swarm algorithm, namely AGPA (assisted living Power Amplifier), wherein each round of optimization comprises a hiring bee searching stage, an observation bee searching stage and a bee detecting searching stage; in the employed bee searching stage, the unloading scheme corresponding to each food source learns the unloading scheme corresponding to another food source selected randomly with random probability to obtain a new task unloading scheme, and the FGRA is used to obtain a corresponding resource allocation scheme and a maximum service experience deficiency coefficient
The superior food source of the two before and after learning is preserved greedily and the history food source is updated; in the observation bee searching stage, the unloading scheme corresponding to each food source learns the unloading scheme corresponding to the historical optimal food source with random probability to obtain a new task unloading scheme, and the corresponding resource allocation scheme and the maximum service experience deficiency coefficient are obtained by using FGRA
The superior food source in the two food sources before and after the study is greedily stored and the historical optimal food source is updated; in the search stage of scouting bees, each food sourceThe corresponding unloading scheme is explored in a random direction with random probability to obtain a new task unloading scheme, and the corresponding resource allocation scheme and the maximum service experience deficiency coefficient are obtained by using FGRA
The superior food source of the two food sources before and after learning is greedily saved and the historical optimal food source is updated; after the Cycle round of optimization, the task unloading and resource allocation scheme corresponding to the historical optimal food source is the task unloading and resource allocation scheme for ensuring the user fairness to be maximized.
In step 1), initializing one food source in the population is specifically: for task collections
In the method, each task randomly selects an edge computing server configured with the service type required by the task as an unloading place, and after a complete unloading scheme is obtained, a corresponding resource allocation scheme and a maximum service experience deficiency coefficient are computed by using FGRA
In step 1), a resource allocation method for guaranteeing user fairness, called FGRA, is used, and the resource allocation method is based on a mechanism of minimizing a maximum service experience deficiency coefficient of a single edge computing server, and uses a binary search method, which specifically includes the following steps:
1.1) initializing a Single edge compute Server McompHas an upper limit of service experience deficiency coefficient of
Service experience deficiency coefficient lower bound
Compute the server M according to the distribution to the edgecompTask set of
Calculating the attribute of (2); wherein the task TiIs given by a service experience deficiency factor giThe calculation method is as follows:
in the formula (d)iFor task TiActual time delay of, DiFor task TiIs arranged and combined into the form described above, and uses the auxiliary parameter thetaiAnd ηiConstant part in alternative equation; as can be seen from the above equation, the edge computing server McompThe lower limit of the service experience deficiency coefficient of the middle task is infinitely close to thetai(ii) a Since theta is not really taken in the actual calculationiThis value, therefore, is set only preliminarily
1.2) order McompHas a maximum service experience deficiency coefficient of
Order to
All tasks of (2) service experience deficiency coefficient Gi＝gcompAnd calculating the resource quantity f required by each task under the service experience deficiency coefficient g according to the following formulacomp,i：
1.3) to edge calculation Server McompThe number of resources of the task of (2) is summed:
if this value is greater than McompTotal computing resources of FcompThe search range of the minimum service experience deficiency coefficient is increased,
if this value is less than McompTotal computing resources of FcompThe search range of the minimum service experience deficiency coefficient is narrowed,
1.4) repeating steps 1.2), 1.3) until the edge calculation server M is reachedcompThe number of resources of the task
Or when the maximum number of iterations is reached, g in this casecompThe value is
The service experience deficiency coefficients of all the tasks are calculated according to the service experience deficiency coefficients
The number of resources allocated to all tasks in the set.
In step 2), the task offloading scheme optimally uses an evolutionary algorithm combining the concepts of the swarm algorithm, the genetic algorithm and the particle swarm algorithm, called AGPA, which includes the following contents:
2.1) hiring bee search phase: learning by each food source in the population to another food source selected randomly with a random probability; i.e. the unloading scheme Γ for the jth food source in the populationjRandomly selecting a different unloading scheme Γ corresponding to the food source kkIf a learning rate L is randomly generated, the learning rate L belongs to (0,0.5), and the number of task learning under the learning rate is
Where n is the task set
The number of tasks in is equal tojIn the method, NL tasks are randomly selected, and the unloading positions of the NL tasks are respectively changed into the positions corresponding to the load of the gammakThe same unloading site is obtained to obtain a learned unloading scheme gamma'j(ii) a Next, using FGRA, Γ is calculatedjCorresponding resource allocation scheme and corresponding maximum service experience deficiency coefficient thereof, greedily storing lower gammajAnd gammajThe superior food source of the two and updates the historical food source;
2.2) observation bee searching stage: learning each food source in the population to the historical optimal food source with random probability, namely, for the unloading scheme gamma corresponding to the jth food source in the populationjThe learning object is the unloading scheme gamma corresponding to the currently explored historical optimal food sourcebest(ii) a Randomly generating a learning rate L epsilon (0,0.5), and then the number of task learning under the learning rate is
Where n is the task set
The number of tasks in is equal tojIn the method, NL tasks are randomly selected, and the unloading positions of the NL tasks are respectively changed into the positions corresponding to the load of the gammabestThe same unloading place is obtained to obtain a learned unloading scheme gammaj(ii) a Next, using FGRA, Γ is calculatedjCorresponding resource allocation scheme and corresponding maximum service experience deficiency coefficient thereof, greedily storing lower gammajAnd gammajThe superior food source of the two and the historical optimal food source is updated;
2.3) searching the scout bees: exploring each food source in the population in random directions with random probability, namely, for the unloading scheme gamma corresponding to the jth food source in the populationjRandomly generating a learning rate L epsilon (0,0.5), and then the number of tasks for random search under the learning rate is
Wherein n is a task setCombination of Chinese herbs
The number of tasks in is equal tojIn the method, NL tasks are randomly selected, another unloading place selected by an edge computing server configured with a service type required by the tasks is randomly selected for the NL tasks, and an unloading scheme gamma after retrieval is obtainedj(ii) a Then using FGRA to calculate at ΓjCorresponding resource allocation scheme and corresponding maximum service experience deficiency coefficient, and greedily storing lower gammajAnd Γ'jThe superior food source of the two and the historical optimal food source is updated;
2.4) repeating the steps 2.1), 2.2) and 2.3) until the maximum iteration number is reached, and after the Cycle round of optimization, the task unloading and resource allocation scheme corresponding to the historical optimal food source is the task unloading and resource allocation scheme for ensuring the user fairness to be maximized.
Compared with the prior art, the invention has the following advantages and beneficial effects:
1. the method of the invention does not depend on the cloud server, reasonably utilizes the computing resources in the edge server cluster with relatively short geographic distance, and improves the utilization rate of the computing resources.
2. The method of the invention optimizes the fairness of the users, ensures the fair treatment of each user during the allocation of the computing resources and optimizes the service experience of the users.
3. The method can simultaneously optimize the time delay and the utilization rate of the computing resources.
Detailed Description
The present invention will be described in further detail with reference to examples, but the embodiments of the present invention are not limited thereto.
The embodiment provides a task unloading and resource allocation method based on user fairness maximization, and one computing server comprising m edges is considered
Region of (1), MmFor the mth edge calculation clothesA server in which n tasks are distributed
TnFor the nth task, each task needs to be unloaded to the edge computing server, and computing is carried out by using computing resources provided by the edge computing server; here, a scenario where m is 16 and n is 50 is assumed; t isiFor the ith task, task TiService experience deficiency coefficient GiDefined as the actual time delay diWith maximum allowable delay DiRatio of (i.e. G)i＝di/DiWherein task TiActual time delay d ofiThe task set is the sum of the transmission and propagation delay of the task and the calculation delay on the edge calculation server
The maximum service experience deficiency factor in is
The method is to minimize
Determining, for each task, an unloading location and an amount of computing resources allocated for the task; therefore, the target function design of the edge computing service platform (the edge computing service platform is a generic term of all edge computing servers, that is, all edge computing servers form the edge computing service platform) is as follows:
wherein gamma is the unloading scheme of the task,
a resource allocation scheme for the task; fcompComputing a server M for an edgecompTotal amount of computing resources of fcomp,iIs McompAssigned to task TiThe number of computing resources of (a); siFor task TiThe type of service that is required is,
computing a server M for an edgecompThe provided service type set, wherein the total number of the service types is assumed to be 8, and each server can only provide 3 services at most; the limiting conditions are, respectively, the edge calculation server McompThe sum of the computing resources allocated to the task does not exceed McompThe total amount of the resources and the task T are calculatediActual time delay d ofiNot exceeding TiMaximum allowable delay DiAnd the offloading site of the task, needs to select the edge compute server that has configured its required service type.
The task unloading and resource allocation method based on user fairness maximization specifically comprises the following steps:
1) initializing a population with NP food sources, wherein each food source corresponds to a task unloading scheme and a corresponding resource allocation scheme; here, each task offload scheme can be represented by a 1 × 10 vector, such as [2,16,4,7,3,3,5,10,6,7], which indicates the number of 10 task offload edge computing servers;
2) the method for acquiring the task unloading scheme needing resource allocation at the moment, using a resource allocation method for guaranteeing user fairness, called FGRA, based on a mechanism of minimizing the maximum service experience deficiency coefficient of a single edge computing server, using a binary search method, and specifically comprising the following steps:
2.1) initializing a Single edge compute Server McompHas an upper limit of service experience deficiency coefficient of
Service experience deficiency coefficient lower bound
According to distribution to the edgeComputation server McompTask set of
Calculating the attribute of (2); wherein the task TiIs given by a service experience deficiency factor giThe calculation method is as follows:
wherein d isiFor task TiActual time delay of, DiFor task TiIs arranged and combined into the form described above, and uses the auxiliary parameter thetaiAnd ηiReplacing the constant part in the equation; as can be seen from the above equation, the edge computing server McompThe lower limit of the service experience deficiency coefficient of the middle task is infinitely close to thetai(ii) a Since theta is not really taken in the actual calculationiThis value, therefore, is set only preliminarily
Here, the parameter characteristics of 10 tasks are calculated
2.2) order McompHas a maximum service experience deficiency coefficient of
Order to
All tasks of (2) service experience deficiency coefficient Gi＝gcompHere, g is calculatedcomp(1+ 0.2)/2-0.6, and calculating the quantity f of resources required by each task under the service experience deficiency coefficient g according to the following formulacomp,i：
2.3) to edge calculation Server McompThe number of resources of the task of (2) is summed:
if this value is greater than McompTotal computing resources of FcompThe search range of the minimum service experience deficiency coefficient is increased,
if this value is less than McompTotal computing resources of FcompThe search range of the minimum service experience deficiency coefficient is narrowed,
here, the method determines
Thereby narrowing the search range of the minimum service experience deficiency coefficient, and enabling
2.4) repeating steps 2.2), 2.3) until the edge calculation server M is reachedcompThe number of resources of the task
Or when the maximum number of iterations is reached, g in this casecompThe value is
The service experience deficiency coefficients of all the tasks are calculated according to the service experience deficiency coefficients
The number of resources allocated to all tasks in the set.
In the FGRA resource allocation algorithm, the following two propositions are proposed and proved that FGRA can achieve the maximum service experience deficiency factor minimization of a single edge computing server.
Proposition 1: the maximum service experience factor of an edge computing server can be minimized only if the service experience factors of all tasks on the edge computing server are equal.
And (3) proving that: if server M is computed at the edgecompHas a task TmTask T with a service experience deficiency factor that is greater than the maximum service experience deficiency factor on the serveriIf it is small, T can be appropriately setmComputing resource of (1) is transferred to (T)iUpper, then TiHas a reduced number of service experience deficiency, i.e. the edge computing server McompThe maximum service experience deficiency factor is reduced. Therefore, the maximum service experience factor of an edge computing server can be minimized if and only if the service experience factors of all tasks on the edge computing server are equal.
Proposition 2: the maximum service experience factor of an edge compute server is minimized only if the compute resources on the edge compute server are fully allocated.
And (3) proving that: if server M is computed at the edgecompIf resources are not allocated to the task, then the resources are allocated to the task T with the maximum service experience deficiency coefficient on the serveriThen T isiHas a reduced number of service experience deficiency, i.e. the edge computing server McompThe maximum service experience deficiency factor of (2) is reduced. Therefore, the maximum service experience factor of an edge computing server can be minimized if and only if the service experience factors of all tasks on the edge computing server are equal.
3) The task unloading scheme optimally uses a evolutionary algorithm combining the ideas of a swarm algorithm, a genetic algorithm and a particle swarm algorithm, is called AGPA, and comprises the following steps:
3.1) hiring bee search phase: learning by each food source in the population to another food source selected randomly with a random probability; i.e. the unloading scheme Γ for the jth food source in the populationjRandomly select one to be associated withUnloading schemes gamma corresponding to different food sources kkIf a learning rate L is randomly generated, the learning rate L belongs to (0,0.5), and the number of task learning under the learning rate is
Where n is the task set
The number of tasks in is equal tojIn the method, NL tasks are randomly selected, and the unloading positions of the NL tasks are respectively changed into the positions corresponding to the load of the gammakThe same unloading site is obtained to obtain a learned unloading scheme gamma'j(ii) a Then, Γ 'is calculated using FGRA'jCorresponding resource allocation scheme and corresponding maximum service experience deficiency coefficient thereof, greedily storing lower gammajAnd Γ'jThe superior food source of the two and updates the historical food source; here, if L is 0.2, the number NL of learned tasks is 0.2 × 10, 2. Suppose ΓjIs [2,16,4,7,3,3,5,10,6,7]]，ΓkIs [3,5,8,6,5,13,16,4,7,9 ]]Then two tasks to learn are randomly selected, and are respectively the third task T3And a seventh task T7Then the learned unloading scheme is [2,16,8,7,3,3,16,10,6,7]. And then the FGRA is used for calculating the corresponding resource allocation scheme, and the learned scheme is found to be more excellent, so the gamma isjThe corresponding scheme is changed to [2,16,8,7,3,3,16,10,6, 7]]。
3.2) observation bee searching stage: learning each food source in the population to the historical optimal food source with random probability, namely, for the unloading scheme gamma corresponding to the jth food source in the populationjThe learning object is the unloading scheme gamma corresponding to the currently explored historical optimal food sourcebest(ii) a Randomly generating a learning rate L epsilon (0,0.5), and then the number of task learning under the learning rate is
Where n is the task set
The number of tasks in is equal tojIn the method, NL tasks are randomly selected, and the unloading positions of the NL tasks are respectively changed into the positions corresponding to the load of the gammabestThe same unloading site is obtained to obtain a learned unloading scheme gamma'j(ii) a Γ 'was then calculated using FGRA'jCorresponding resource allocation scheme and corresponding maximum service experience deficiency coefficient thereof, greedily storing lower gammajAnd Γ'jThe superior food source of the two and the historical optimal food source is updated; the steps are similar to 3.1);
3.3) searching the scout bees: exploring each food source in the population in random directions with random probability, namely, for the unloading scheme gamma corresponding to the jth food source in the populationjRandomly generating a learning rate L epsilon (0,0.5), and then the number of tasks for random search under the learning rate is
Wherein n is a task set
The number of tasks in is equal tojIn the method, NL tasks are randomly selected, another unloading place selected by an edge computing server configured with service types required by the tasks is randomly selected for the NL tasks, and an unloading scheme Γ 'after search is obtained'j(ii) a Then, using FGRA, Γ 'is calculated'jCorresponding resource allocation scheme and corresponding maximum service experience deficiency coefficient, and greedily storing lower gammajAnd Γ'jThe superior food source of the two and the historical optimal food source is updated; where L is 0.35, the number of learned tasks
Suppose ΓjIs [2,16,4,7,3,3,5,10,6,7]]Then, 4 tasks to be explored are randomly selected, which are the third, fifth, eighth and tenth tasks T3、T5、T8And T10Then the unloading scheme after searching is [2,16,8,7,9,3,5,9,6 ]]. Calculating pairs by using FGRA methodThe scheme before searching is found to be more excellent according to the resource allocation scheme, so that the gamma isjThe corresponding scheme is changed to [2,16,4,7,3,3,5,10,6,7]]。
And 3.4) repeating the steps 3.1), 3.2) and 3.3) until the maximum iteration number is reached, and after the optimization of the Cycle wheel, the task unloading and resource allocation scheme corresponding to the historical optimal food source is the task unloading and resource allocation scheme for ensuring the user fairness to be maximized.
The above embodiments are preferred embodiments of the present invention, but the present invention is not limited to the above embodiments, and any other changes, modifications, substitutions, combinations, and simplifications which do not depart from the spirit and principle of the present invention should be construed as equivalents thereof, and all such changes, modifications, substitutions, combinations, and simplifications are intended to be included in the scope of the present invention.
Claims (5)
1. A task unloading and resource allocation method based on user fairness maximization considers a computing server comprising m edges
Region of (1), MmFor the mth edge, a server is calculated in which n tasks are distributed in the region
TnFor the nth task, each task needs to be unloaded to the edge computing server, and computing is carried out by using computing resources provided by the edge computing server; t isiFor the ith task, task TiService experience deficiency coefficient GiDefined as the actual time delay diWith maximum allowable delay DiRatio of (i.e. G)i＝di/DiWherein task TiActual time delay d ofiThe task set is the sum of the transmission and propagation delay of the task and the calculation delay on the edge calculation server
Maximum service experience owe inHas a defect coefficient of
Wherein the method is characterized by minimizing the set of tasks
Maximum service experience deficiency factor in
For the purpose of determining the unloading place and the amount of the distributed computing resources for each task, an objective function of the edge computing service platform is designed as follows:
wherein gamma is the unloading scheme of the task,
resource allocation schemes for tasks, FcompComputing a server M for an edgecompTotal amount of computing resources of fcomp,iIs McompAssigned to task TiNumber of computing resources of SiFor task TiThe type of service that is required is,
is McompA set of offered service types; the limiting conditions are respectively as follows: mcompThe sum of the computing resources allocated to the task does not exceed McompTotal number of computing resources, task TiActual time delay d ofiNot exceeding TiMaximum allowable delay DiAnd the unloading place of the task needs to select the edge computing server configured with the required service type;
the edge computing service platform is a general name of all edge computing servers, namely the edge computing service platform is formed by all the edge computing servers.
2. The method of claim 1, comprising the following steps:
1) initializing a population having NP food sources, wherein each food source corresponds to a task offloading scheme and a corresponding resource allocation scheme; then, a resource allocation method for guaranteeing user fairness, called FGRA, is used, which calculates for each task offload solution the corresponding resource allocation solution and its corresponding maximum service experience deficiency factor
2) Performing Cycle round optimization on food sources in a population by using an evolutionary algorithm combining the ideas of a swarm algorithm, a genetic algorithm and a particle swarm algorithm, namely AGPA (accelerated g particle swarm optimization), wherein each round of optimization comprises a hired bee searching stage, an observed bee searching stage and a bee detecting searching stage; in the stage of employing bee search, the unloading scheme corresponding to each food source learns the unloading scheme corresponding to another food source selected randomly with random probability to obtain a new task unloading scheme, and uses FGRA to obtain the corresponding resource allocation scheme and maximum service experience deficiency coefficient
The superior food source of the two food sources before and after learning is greedily stored and the historical food source is updated; in the observation bee searching stage, the unloading scheme corresponding to each food source learns the unloading scheme corresponding to the historical optimal food source with random probability to obtain a new task unloading scheme, and the corresponding resource allocation scheme and the maximum service experience deficiency coefficient are obtained by using FGRA
The superior food source of the two food sources before and after learning is greedily saved and the historical optimal food source is updated; under investigationIn the bee searching stage, the unloading scheme corresponding to each food source is explored in a random direction with random probability to obtain a new task unloading scheme, and the corresponding resource allocation scheme and the maximum service experience deficiency coefficient are obtained by using FGRA
The superior food source of the two food sources before and after learning is greedily saved and the historical optimal food source is updated; after the Cycle round of optimization, the task unloading and resource allocation scheme corresponding to the historical optimal food source is the task unloading and resource allocation scheme for ensuring the user fairness to be maximized.
3. The method for task offloading and resource allocation based on user fairness maximization of claim 2, wherein in step 1), initializing one food source in the population is: for task collections
In the method, each task randomly selects an edge computing server configured with the service type required by the task as an unloading place, and after a complete unloading scheme is obtained, a corresponding resource allocation scheme and a maximum service experience deficiency coefficient are computed by using FGRA
4. The method for task offloading and resource allocation based on user fairness maximization as claimed in claim 2, wherein in step 1), a resource allocation method for guaranteeing user fairness, called FGRA, is used, and the resource allocation method is based on a mechanism for minimizing maximum service experience deficiency factor of a single edge computing server, and uses a binary search method, which specifically includes the following steps:
1.1) initializing a Single edge compute Server McompHas an upper limit of service experience deficiency coefficient of
Service experience deficiency coefficient lower bound
Compute the server M according to the distribution to the edgecompTask set of
Calculating the attribute of (2); wherein the task TiIs given by a service experience deficiency factor giThe calculation method is as follows:
in the formula (d)iFor task TiActual time delay of, DiFor task TiIs arranged and combined into the form described above, and uses the auxiliary parameter thetaiAnd ηiConstant part in alternative equation; as can be seen from the above equation, the edge computing server McompThe lower limit of the service experience deficiency coefficient of the middle task is infinitely close to thetai(ii) a Since theta is not really taken in the actual calculationiThis value, therefore, is set only preliminarily
1.2) order McompHas a maximum service experience deficiency coefficient of
Order to
All tasks of (2) service experience deficiency coefficient Gi＝gcompAnd calculating the resource quantity f required by each task under the service experience deficiency coefficient g according to the following formulacomp,i：
1.3) to edge calculation Server McompThe number of resources of the task of (2) is summed:
if this value is greater than McompTotal computing resources of FcompThe search range of the minimum service experience deficiency coefficient is increased,
if this value is less than McompTotal computing resources of FcompThe search range of the minimum service experience deficiency coefficient is narrowed,
1.4) repeating steps 1.2), 1.3) until the edge calculation server M is reachedcompThe number of resources of the task
Or when the maximum number of iterations is reached, g in this casecompThe value is
The service experience deficiency coefficients of all the tasks are calculated according to the service experience deficiency coefficients
The number of resources allocated to all tasks in the set.
5. The method for task offloading and resource allocation based on user fairness maximization according to claim 2, wherein in step 2), the task offloading scheme optimizes and uses an evolutionary algorithm combining concepts of bee colony algorithm, genetic algorithm and particle swarm algorithm, called AGPA, which comprises the following contents:
2.1) hiring bee search phase: learning by each food source in the population to another food source selected randomly with a random probability; i.e. the unloading scheme Γ for the jth food source in the populationjRandomly selecting an unloading scheme gamma corresponding to a different food source kkIf a learning rate L is randomly generated, the learning rate L belongs to (0,0.5), and the number of task learning under the learning rate is
Where n is the task set
The number of tasks in is equal tojIn the method, NL tasks are randomly selected, and the unloading positions of the NL tasks are respectively changed into the positions corresponding to the load of the gammakThe same unloading site is obtained to obtain a learned unloading scheme gamma'j(ii) a Then, Γ 'is calculated using FGRA'jCorresponding resource allocation scheme and corresponding maximum service experience deficiency coefficient thereof, greedily storing lower gammajAnd Γ'jThe superior food source of the two and updates the historical food source;
2.2) observation bee searching stage: learning each food source in the population to the historical optimal food source with random probability, namely, for the unloading scheme gamma corresponding to the jth food source in the populationjThe learning object is the unloading scheme gamma corresponding to the currently explored historical optimal food sourcebest(ii) a Randomly generating a learning rate L epsilon (0,0.5), and then the number of task learning under the learning rate is
Where n is the task set
The number of tasks in is equal tojIn the random selection of NLThe task changes the load-off positions of the NL tasks to be equal to the load-off positions of the Γ tasksbestThe same unloading site is obtained to obtain a learned unloading scheme gamma'j(ii) a Then, Γ 'is calculated using FGRA'jCorresponding resource allocation scheme and corresponding maximum service experience deficiency coefficient thereof, greedily storing lower gammajAnd Γ'jThe superior food source of the two and the historical optimal food source is updated;
2.3) searching the scout bees: exploring each food source in the population in random directions with random probability, namely, for the unloading scheme gamma corresponding to the jth food source in the populationjRandomly generating a learning rate L epsilon (0,0.5), and then the number of tasks for random search under the learning rate is
Where n is the task set
The number of tasks in is equal tojIn the method, NL tasks are randomly selected, another unloading place selected by an edge computing server configured with service types required by the tasks is randomly selected for the NL tasks, and the explored unloading scheme Γ'j(ii) a Then, using FGRA, Γ 'is calculated'jCorresponding resource allocation scheme and corresponding maximum service experience deficiency coefficient thereof, and greedily storing lower gammajAnd Γ'jThe superior food source of the two and the historical optimal food source is updated;
2.4) repeating the steps 2.1), 2.2) and 2.3) until the maximum iteration number is reached, and after the Cycle round of optimization, the task unloading and resource allocation scheme corresponding to the historical optimal food source is the task unloading and resource allocation scheme for ensuring the fairness of the user to be maximized.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111325086.1A CN114338662B (en) | 2021-11-10 | 2021-11-10 | Task unloading and resource allocation method based on user fairness maximization |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111325086.1A CN114338662B (en) | 2021-11-10 | 2021-11-10 | Task unloading and resource allocation method based on user fairness maximization |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114338662A true CN114338662A (en) | 2022-04-12 |
| CN114338662B CN114338662B (en) | 2022-10-25 |
Family
ID=81044497
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111325086.1A Active CN114338662B (en) | 2021-11-10 | 2021-11-10 | Task unloading and resource allocation method based on user fairness maximization |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114338662B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111586762A (en) * | 2020-04-29 | 2020-08-25 | 重庆邮电大学 | Task unloading and resource allocation joint optimization method based on edge cooperation |
| CN111953759A (en) * | 2020-08-04 | 2020-11-17 | 国网河南省电力公司信息通信公司 | Collaborative computing task unloading and transferring method and device based on reinforcement learning |
| US20210266834A1 (en) * | 2020-02-25 | 2021-08-26 | South China University Of Technology | METHOD OF MULTI-ACCESS EDGE COMPUTING TASK OFFLOADING BASED ON D2D IN INTERNET OF VEHICLES (IoV) ENVIRONMENT |
-
2021
- 2021-11-10 CN CN202111325086.1A patent/CN114338662B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210266834A1 (en) * | 2020-02-25 | 2021-08-26 | South China University Of Technology | METHOD OF MULTI-ACCESS EDGE COMPUTING TASK OFFLOADING BASED ON D2D IN INTERNET OF VEHICLES (IoV) ENVIRONMENT |
| CN111586762A (en) * | 2020-04-29 | 2020-08-25 | 重庆邮电大学 | Task unloading and resource allocation joint optimization method based on edge cooperation |
| CN111953759A (en) * | 2020-08-04 | 2020-11-17 | 国网河南省电力公司信息通信公司 | Collaborative computing task unloading and transferring method and device based on reinforcement learning |
Non-Patent Citations (4)
| Title |
|---|
| WEN JIANG FENG,ET AL.: "Multi-User_and_Multi-Task_Offloading_Decision_Algorithms_Based_on_Imbalanced_Edge_Cloud", 《IEEE》 * |
| 杨天等: "MEC中卸载决策与资源分配的深度强化学习方法", 《计算机工程》 * |
| 闫伟等: "基于自适应遗传算法的MEC任务卸载及资源分配", 《电子技术应用》 * |
| 黄晓舸等: "基于MEC的任务卸载和资源分配联合优化方案", 《系统工程与电子技术》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114338662B (en) | 2022-10-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109561148B (en) | Distributed task scheduling method based on directed acyclic graph in edge computing network | |
| CN111163519B (en) | Wireless body area network resource allocation and task offloading method with maximized system benefit | |
| CN112512056B (en) | Multi-objective optimization calculation unloading method in mobile edge calculation network | |
| CN110351754B (en) | Industrial Internet machine equipment user data calculation unloading decision method based on Q-learning | |
| CN111262944B (en) | Method and system for hierarchical task offloading in heterogeneous mobile edge computing network | |
| Wang et al. | Reputation-enabled federated learning model aggregation in mobile platforms | |
| CN111836284B (en) | Energy consumption optimization calculation and unloading method and system based on mobile edge calculation | |
| CN108830408A (en) | A kind of equipment support task multi-objective planning method that oriented mission is resource matched | |
| CN114625506A (en) | Edge cloud collaborative task unloading method based on adaptive covariance matrix evolution strategy | |
| CN111047040A (en) | Web service combination method based on IFPA algorithm | |
| Zhang et al. | Quantified edge server placement with quantum encoding in internet of vehicles | |
| CN113139639B (en) | MOMBI-oriented smart city application multi-target computing migration method and device | |
| CN109840625B (en) | Courier group path navigation method | |
| CN114219094A (en) | Communication cost and model robustness optimization method based on multi-task federal learning | |
| CN114338662B (en) | Task unloading and resource allocation method based on user fairness maximization | |
| CN117669741A (en) | Unmanned aerial vehicle cluster size model dynamic collaborative reasoning method based on genetic algorithm | |
| CN110635486B (en) | Load frequency modulation optimal scheduling method considering constraint conditions of power distribution network | |
| CN116613733A (en) | Multi-element energy system optimal scheduling method and device considering uncertainty of source load | |
| CN117196884A (en) | Virtual power plant optimization control method and system | |
| CN116321189A (en) | Server deployment method based on deep reinforcement learning in edge-oriented computing | |
| CN113810916B (en) | Multi-server mixed deployment architecture and method in 5G/6G edge computing scene | |
| CN115766478A (en) | Unloading method of air-ground cooperative edge computing server | |
| CN111640043B (en) | Power distribution network reconstruction method and device, computer equipment and storage medium | |
| CN115361453A (en) | Load fair unloading and transferring method for edge service network | |
| CN113378481A (en) | Internet data center demand response optimization method based on multi-objective evolutionary algorithm |
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 |