CN113950103B - Multi-server complete computing unloading method and system under mobile edge environment - Google Patents
Multi-server complete computing unloading method and system under mobile edge environment Download PDFInfo
- Publication number
- CN113950103B CN113950103B CN202111061668.3A CN202111061668A CN113950103B CN 113950103 B CN113950103 B CN 113950103B CN 202111061668 A CN202111061668 A CN 202111061668A CN 113950103 B CN113950103 B CN 113950103B
- Authority
- CN
- China
- Prior art keywords
- task
- server
- unloading
- mobile
- mec
- 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.)
- Active
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/02—Traffic management, e.g. flow control or congestion control
- H04W28/08—Load balancing or load distribution
- H04W28/09—Management thereof
- H04W28/0925—Management thereof using policies
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W24/00—Supervisory, monitoring or testing arrangements
- H04W24/06—Testing, supervising or monitoring using simulated traffic
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04W—WIRELESS COMMUNICATION NETWORKS
- H04W28/00—Network traffic management; Network resource management
- H04W28/02—Traffic management, e.g. flow control or congestion control
- H04W28/08—Load balancing or load distribution
- H04W28/09—Management thereof
- H04W28/0958—Management thereof based on metrics or performance parameters
-
- 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
- Y02D30/00—Reducing energy consumption in communication networks
- Y02D30/70—Reducing energy consumption in communication networks in wireless communication networks
Abstract
The invention belongs to the technical field of task unloading and resource allocation of mobile edge computing, and discloses a method and a system for unloading multi-server complete computing in a mobile edge environment, wherein the method for unloading multi-server complete computing in the mobile edge environment comprises the following steps: constructing an application scene; designing a network communication model; constructing a task computing model; planning a problem of a multi-user multi-server MEC system; and a complete calculation unloading scheme based on near-end strategy optimization is provided, and the dynamic target optimization problem of the multi-user multi-MEC server system is solved. Compared with the experiment by adopting four standard algorithms of local execution, polling scheduling, random unloading and complete calculation unloading based on DQN, the experiment measures the total benefit of the system, the task execution delay and the energy consumption of the mobile equipment aiming at different parameters such as the calculation capacity of the server, the number of terminal equipment, the number of MEC servers and the like, realizes higher long-term total benefit of the system and improves the user experience quality.
Description
Technical Field
The invention belongs to the technical field of task unloading and resource allocation of mobile edge computing, and particularly relates to a multi-server complete computing unloading method and system in a mobile edge environment.
Background
At present, due to the continuous popularization of the internet of things technology and the rapid development of the mobile communication field, the construction of the 5G core network architecture faces more challenges. Mobile Edge Computing (MEC) is a core technology of 5G, and in this field, computing offloading technology has become a hot spot direction, and whether a task needs to be offloaded, the size of an offload amount, and the time for executing task offloading are factors that need to be considered in making task offloading decisions. In addition, in a multi-user computing offload scenario, it is also an important challenge to achieve reasonable allocation of base station wireless bandwidth resources and server computing resources. Therefore, how to make the most reasonable task offloading decision and resource allocation scheme according to the terminal device and the surrounding environment factors is a difficult problem to be solved in the current computing offloading framework research, and a large number of task offloading strategies based on different algorithms and optimization targets are also provided. According to the optimization target of the algorithm, the calculation unloading strategy can be divided into the following three types:
the first category of strategies is mainly to reduce task processing delays. For example, wang and Yao et al propose a multi-layer data stream processing system composed of a central cloud, an MEC server, and edge devices in "oil task assignment, transmission, and computing resource allocation in multilayer mobile computing systems" (IEEE Internet of threads Journal,2019,6 (2): 2872-2884.), and realize minimization of task processing delay by considering the problems of joint task scheduling and transmission and computing resource allocation in the system. The disadvantages of the method are: the energy consumption of one end of the mobile terminal equipment during calculation unloading cannot be considered, and the terminal equipment may have the situation that the unloading strategy cannot normally operate due to insufficient electric energy.
The second category of strategies is mainly to reduce the device energy consumption. For example, in "Optimal Mobile computing Offloading with Hard delay Constraints" (IEEE Transactions on Mobile computing,2020,19 (9): 2160-2173) by a.hekmani and p.teymoori et al, the Computation Offloading problem constrained by the task completion Deadline is considered, and an online energy optimization Computation Offloading algorithm allowing remote Offloading and local tasks to be performed simultaneously is studied to obtain the minimum average Mobile equipment energy under a uniform markov wireless channel under the condition of meeting the task completion Deadline. The method has the following defects: in some systems, users prefer to minimize the sum of the time consumption and the total energy consumption of the system to reduce the overall consumption of the system or to balance the time consumption and the energy consumption so that the total consumption of the system is in a relatively optimal and stable state.
The third type of strategy is mainly time delay and energy consumption compromise. In "Task off flow and Resource Allocation for Mobile Edge Computing by Deep discovery requirement on SARSA" (IEEE Access,2020, 8. The disadvantages of the method are: different systems may have different performance requirements during actual offloading rather than being limited to latency and energy consumption.
In summary, a new method for completely computing and offloading multiple servers in a mobile edge environment is needed to make up for the defects of the existing computing and offloading strategies, reduce the time delay, and reduce the energy consumption.
Through the above analysis, the problems and defects of the prior art are as follows:
(1) In the existing calculation unloading strategy for reducing task processing delay, energy consumption at one end of a mobile terminal device during calculation unloading cannot be considered, and the terminal device may have the situation that the unloading strategy cannot normally run due to insufficient electric energy; meanwhile, the total time delay of the task execution of the existing model is long, and the total energy consumption of the mobile equipment is high.
(2) In the existing calculation unloading strategy for reducing the energy consumption of equipment, in some systems, a user more hopes that the sum of the time consumption and the total energy consumption of the system can be minimized to reduce the overall consumption of the system or balance the time consumption and the energy consumption, so that the total consumption of the system is in a relatively better and stable state.
(3) In the existing calculation unloading strategy with compromise between time delay and energy consumption, different systems may have different performance requirements in the actual unloading process rather than being limited to time delay and energy consumption.
The difficulty in solving the above problems and defects is: in various current IoT scenes, how to make the most reasonable task offloading decision and resource allocation scheme according to the terminal device and the surrounding environment factors is an urgent problem to be solved in the research of the computational offloading framework, and the complex environment causes longer task delay and higher total energy consumption of the device.
The significance of solving the problems and the defects is as follows: under the more complex environment of multiple mobile devices and multiple MEC servers, a complete computation unloading scheme taking the total task execution time delay and the total mobile device energy consumption as the common optimization indexes is urgently needed.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a method and a system for multi-server complete computation unloading in a mobile edge environment.
The invention is realized in such a way that a method for unloading multi-server complete computation in a mobile edge environment comprises the following steps:
constructing an application scene with a multi-user terminal, a single base station and a multi-MEC server, and realizing the unloading decision of the task of the mobile terminal equipment;
step two, calculating a data transmission rate according to a Shannon-Hartley law, constructing a wireless network communication model, and realizing a network bandwidth resource allocation decision of all mobile terminals;
step three, constructing a task calculation model, realizing the analysis of time delay and energy consumption under the condition of processing two tasks, and providing basic content for the planning of system problems;
expressing the complete task unloading problem under the multi-user multi-MEC server model in each time slice, realizing the problem planning of the multi-user multi-server MEC system, and providing an optimization target and related constraints for the unloading scheme;
and step five, providing a complete calculation unloading scheme based on near-end strategy optimization, solving the dynamic target optimization problem of the multi-user multi-MEC server system, and realizing the target of minimizing task delay and energy consumption.
Further, in the first step, the constructing an application scenario with a multi-user terminal, a single base station and a multi-MEC server deployed to implement the offloading decision of the task of the mobile terminal device includes:
(1) Constructing application scenarios of MEC system
Under the system, a multi-user terminal and a multi-MEC server are expressed as follows:
U＝{1,2,...}，C＝{1,2,...}；
each terminal u generates an undetachable resource-intensive task T u ＝{q u ,X u ,δ u U is equal to U, wherein q is equal to u Indicating that terminal u will be tasked with T u Uploading the required input data volume to the selected MEC server from the local; x u Representing the workload, namely the required CPU calculation period number; delta u Representing a task T u And (4) cutoff time.
(2) Task offload decision making
The MEC system model divides the total time into time slices with all time lengths, and a new task T is arranged on each time slice T terminal u u To arrive, it needs to be right for task T u Carrying out unloading decision; wherein, the unloading situation comprises two types, one is executed locally, and the other is unloaded to the MEC server for processing, and a binary variable x is defined u E {0,1} is used to denote both cases, where x u =0 represents task T u The calculation is carried out locally by utilizing a local CPU; x is the number of u =1 represents task T u Unloading to an MEC server for processing;
import offload decision x = [ x ] 1 ,x 2 ,...,x U ]And y = [ y = 1 ,y 2 ,...,y U ]If task T u Offloaded, defining a vector for each mobile user u:
y u ＝[y u,1 ,y u,2 ,...,y u,c ,...,y u,C ]；
wherein, y u,c ∈{0,1}，y u,c =1 denotes selecting the task T u Off-loading to edge server c; representing a task T u Not offloaded to edge server c; task T u Can and can only be offloaded to one edge server, i.e. the decision variable y u,c The following constraints are satisfied:
(3) Setting and updating a task buffer queue:
each user terminal u is provided with a local waiting task buffer queue I which is calculated in a first-in first-out FIFO sequence u If user u chooses to execute task T locally u Add it to the calculation queue I u Tail, K u Is queue I u The amount of tasks to be performed, the dynamic update formula is expressed as:
K u (t+1)＝max{K u (t)+(1-x u (t))X u (t)-φ u (t),0},u∈U；
wherein, K u (t + 1) the amount of tasks to be performed, x, on the queue is calculated for end user u on time slice t +1 u To indicate task T u Decision variables for local processing or edge offload, X u Is to execute task T u The number of CPU cycles required, phi u (t) represents the local computational workload of the mobile device u within the time slice t.
(4) Maintaining and updating task computation queues:
each MEC server is responsible for maintaining | U | task computation queues λ c ＝[λ 1,c ,λ 2,c ,...,λ u,c ,...,λ |U|,c ]And each queue λ u,c Corresponding to a mobile terminal u; if the edge server receives the unloading task T within the time slice T u Put the task T u Corresponding queue lambda of incoming MEC server u,c In, the update formula is:
wherein M is u,c (t + 1) represents the amount of tasks to be executed in the task queue maintained by the edge server c for the terminal u in the time slice t +1, y u,c To indicate task T u The decision variable whether or not to be offloaded to server c,
then it is the calculation queue λ within time slice t u,c The amount of tasks performed.
Further, in the second step, the step of calculating the data transmission rate according to the Shannon-Hartley law, constructing a wireless network communication model, and implementing a network bandwidth resource allocation decision for all mobile terminals includes:
in the network communication model, assuming that the total available bandwidth of a wireless channel in the wireless network communication model in which a plurality of user equipment transmit on orthogonal channels is B, the uplink data transmission rate between the tth time slice mobile user u and the MEC server c is represented by r u,c It is shown that the network bandwidth resource allocation decision made for all mobile terminals over time slice t is represented as:
b＝[b 1 ,b 2 ,...,b u ,...,b U ]；
wherein, b u Ratio of bandwidth resources, P, allocated to the u mobile terminal by the base station for the t time slice u For the data transmission power of the mobile terminal u, N 0 Is the Gaussian white noise channel power, h u,c Representing the channel gain between the t time slice user equipment u and the edge server c, the uplink data transmission rate r between the t time slice mobile user u and the MEC server c u,c The calculation formula is as follows:
proportion b of network uplink bandwidth resources allocated to each user terminal in time slice t u The following constraints are satisfied:
further, in step three, the constructing of the task calculation model realizes the analysis of time delay and energy consumption under the two task processing situations, and provides basic contents for system problem planning, including:
in the task unloading model, a user u carries a task T u Is made of a variable x u And y u,c Decide together, so analyze task T separately u In the local task queue I u Processing and offloading to a corresponding queue λ on edge server c u,c The time delay and energy consumption of execution are divided into the following steps according to two situations:
(1) Building a complete local computation model
When unloading variable x u =0, i.e. selecting task T u Adding into a local waiting processed task buffer queue I u Time of day, it calculates the time delay locally
Expressed as:
wherein, X u Representing a processing task T u The number of cycles required for the CPU to count,
is the local CPU computation power of end user u. In addition, task T u Waiting for local buffer queue I u The above tasks can be executed after all the tasks are processed, and the waiting time of the process is calculated as:
thus task T u The local total service delay of (a) is expressed as follows:
task T u Selecting energy consumption when executing locally
According to dynamic voltage scaling techniquesBy computing task T u Working load and local CPU working frequency of
It can be calculated as:
where κ is an effective switched capacitance parameter that depends on the chip structure.
(2) Constructing MEC server offload computation model
When unloading variable x u 1 and y u,c When =1, i.e. task T u Is decided to be offloaded to the c-th MEC server. The terminal device u sends the task T u Time delay of unloading to c MEC server through base station
And energy consumption
Respectively as follows:
wherein, y u,c Is to indicate task T u Decision variable, q, whether to offload to Server c u Representing a task T u Size of the uploaded data amount r u,c For the uplink transmission rate between device u and server c, obtained according to the above network communication model, P u Is the transmit power of the u-th user terminal.
Given task queue lambda u,c Is calculated by the computing resource ratio allocation of F u,c Then task T u Execution time on edge server c
Comprises the following steps:
wherein f is c MEC Representing the CPU computing power of the MEC server c.
Task T u Corresponding task queue lambda of the c-th MEC server u,c Waiting time of
Comprises the following steps:
task T on end user u under edge offload u Total service delay of
The calculation is as follows:
the total energy consumption of the end user u is the task T unloaded to the edge server c u The transmission energy consumption of (a), namely:
further, in step four, the expression of the complete task unloading problem under the multi-user multi-MEC server model in each time slice realizes problem planning of the multi-user multi-server MEC system, and provides an optimization target and related constraints for the unloading scheme, including:
the multi-user multi-server task computing model is established as follows:
wherein the content of the first and second substances,
and
respectively representing tasks T generated by the u th mobile device of the current time slice u The execution latency of the full local computation and full offload to the server side,
and
respectively represent tasks T u The transmission energy consumption of the mobile device u under the conditions of local execution energy consumption and edge unloading is introduced, and a parameter mu is introduced 1 And mu 2 To respectively express the weight of time and energy consumption cost and satisfy mu 1 +μ 2 ＝1；
Wherein C1 in the constraint condition represents a task T u Whether to unload to the server side for execution; c2 denotes indicating task T u A variable whether it is offloaded to server c; c3 and C4 respectively indicate that the bandwidth resource ratio allocated to the terminal user by the base station refers to the bandwidth resource ratio allocated to the terminal user by the base station and the calculation resource ratio allocated to the terminal user by the MEC server, and both are not more than 1; c5 indicates that the processing latency of the task does not exceed a given maximum deadline.
Further, in the fifth step, the proposed complete computation offload scheme based on near-end policy optimization solves the dynamic objective optimization problem of the multi-user multi-MEC server system, and achieves the objective of minimizing task delay and energy consumption, including:
(1) Markov decision plan description
On the basis of the reinforcement learning framework, the state, the action and the reward function of the multi-user multi-server MEC system in the time slice t are respectively defined in detail as follows:
1) State: the state S (t) ∈ S of the MDP on the time slice t consists of six parts, respectively the data volume size per mobile device task D (t) = [ D) = 1 (t),D 2 (t),...,D |U| (t)]The number of computing resources required to complete a task X (t) = [ X = [) 1 (t),X 2 (t),...,X |U| (t)]The amount of tasks to be executed on each mobile equipment calculation queue K (t) = [ K = [) 1 (t),K 2 (t),...,K |U| (t)]The method comprises the following steps that an MEC server task quantity matrix M (t) to be executed and an uplink transmission rate matrix r (t), wherein the sizes of the M (t) matrix and the r (t) matrix are | U | × | C |, and elements in the matrix are M respectively u,c And r u,c And representing the amount of tasks to be executed by the task queue maintained by the MEC server c for the mobile device u and the uplink transmission rate between the device u and the MEC server c, the state s (t) is described as s (t) = [ D (t), X (t), K (t), M (t), r (t)]；
2) Action: the action space A of the MDP includes three partial decisions, respectively for task T u Made offload decision x (t) = [ x = [) 1 (t),x 2 (t),...,x| U |(t)]And y (t) = [ y = u,c (t)]U ∈ U, C ∈ C, the network bandwidth resource proportion allocation decision b (t) = [ b ] made by each mobile user 1 (t),b 2 (t),...,b |U| (t)]And the computation resource proportion allocation decision F (t) = [ F ] made by the edge server for all mobile users u,c (t)]U ∈ U, C ∈ C, wherein F ∈ C u,c (t) represents the proportion of computing resources allocated by the MEC server c to the user u over the time slice t, then the action a (t) over the time slice t is defined as a (t) = [ x (t), y (t), b (t), F (t)]；
3) Reward function reward: defining the reward function R (t) as a weighted sum of the calculated delay and energy consumption of all mobile device generated tasks per time slice t, expressed as the target value given by the formula
The goal of the algorithm is to maximize long-term rewards
(2) Optimizing with near-end policy optimization (PPO) algorithm
1) PPO algorithm target
The goal of the PPO algorithm is to maximize the objective function L constrained by the policy update size:
wherein the content of the first and second substances,
the probability ratio between the new strategy and the old strategy is obtained; meanwhile, the PPO algorithm introduces a clip function in the objective function to control the value of γ (θ), which is expressed as follows:
wherein epsilon is a hyper-parameter, the clip function is used for limiting the probability ratio gamma (theta) within the range of [ 1-epsilon, 1+ epsilon ], and the size of strategy updating is effectively adjusted in a mode of selecting the minimum value of the clipped target and the unclipped target;
2) PPO-based complete calculation unloading process
(1) Input task request set T 1 ,T 2 ,...,T U B, total bandwidth of wireless channel, computing power f of each MEC server c MEC ；
(2) Calculating a dominance estimate
Using a combination of pi θ Updating old policy π old ；
(3) Repeating the step (2) for N times, wherein N is the number of state conversion data in each epsilon;
(4) optimizing an objective function L by using w epochs and a small batch size sigma < = N eta;
(5) updating network parameter theta with theta according to gradient descent method old ；
(6) Repeating the step (2) to the step (5) for G times, wherein G is the number of epsilon;
(7) and (3) outputting: task unloading decision x and y, wireless bandwidth resource allocation decision b, and MEC server computing resource allocation decision F.
Another object of the present invention is to provide a system for completely offloading multi-server computing in a mobile edge environment, which applies the method for completely offloading multi-server computing in a mobile edge environment, the system comprising:
the application scene construction module is used for constructing an application scene with a multi-user terminal, a single base station and a multi-MEC server, and realizing the unloading decision of the tasks of the mobile terminal equipment;
the network communication module is used for calculating the data transmission rate according to the Shannon-Hartley law, constructing a wireless network communication model and realizing the network bandwidth resource allocation decision of all the mobile terminals;
the task computing module is used for constructing a task computing model, realizing the analysis of time delay and energy consumption under the condition of processing two tasks and providing basic content for the planning of system problems;
the problem planning module is used for expressing the complete task unloading problem under the multi-user multi-MEC server model in each time slice, realizing the problem planning of the multi-user multi-server MEC system and providing an optimization target and related constraints for the unloading scheme;
and the algorithm optimization module is used for providing a complete calculation unloading scheme based on near-end strategy optimization, solving the dynamic target optimization problem of the multi-user multi-MEC server system and realizing the target of minimizing task delay and energy consumption.
It is a further object of the invention to provide a computer device comprising a memory and a processor, the memory storing a computer program which, when executed by the processor, causes the processor to perform the steps of:
constructing an application scene with a plurality of user terminals, a single base station and a plurality of MEC servers, and realizing the unloading decision of the tasks of the mobile terminal equipment; calculating a data transmission rate according to a Shannon-Hartley law, constructing a wireless network communication model, and realizing a network bandwidth resource allocation decision for all mobile terminals; constructing a task calculation model, realizing the analysis of time delay and energy consumption under the two task processing conditions, and providing basic content for system problem planning; expressing the complete task unloading problem under the multi-user multi-MEC server model in each time slice, realizing the problem planning of a multi-user multi-server MEC system, and providing an optimization target and related constraints for an unloading scheme; and a complete calculation unloading scheme based on near-end strategy optimization is provided, the dynamic target optimization problem of the multi-user multi-MEC server system is solved, and the target of minimizing task delay and energy consumption is realized.
It is another object of the present invention to provide a computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to perform the steps of:
constructing an application scene with a plurality of user terminals, a single base station and a plurality of MEC servers, and realizing the unloading decision of the tasks of the mobile terminal equipment; calculating a data transmission rate according to the Shannon-Hartley law, constructing a wireless network communication model, and realizing a network bandwidth resource allocation decision of all mobile terminals; constructing a task calculation model, realizing the analysis of time delay and energy consumption under the two task processing conditions, and providing basic content for system problem planning; expressing the complete task unloading problem under the multi-user multi-MEC server model in each time slice, realizing the problem planning of a multi-user multi-server MEC system, and providing an optimization target and related constraints for an unloading scheme; and a complete calculation unloading scheme based on near-end strategy optimization is provided, the dynamic target optimization problem of the multi-user multi-MEC server system is solved, and the target of minimizing task delay and energy consumption is realized.
Another object of the present invention is to provide an information data processing terminal, which is used for implementing the multi-server complete computing offloading system in the mobile edge environment.
By combining all the technical schemes, the invention has the advantages and positive effects that: the invention provides a multi-server complete computation unloading method under a mobile edge environment, which constructs an application scene with a multi-user terminal, a single base station and a plurality of MEC servers; designing a network communication model, and making network bandwidth resource allocation decisions for all mobile terminals on a time slice; constructing a task computing model, and discussing time delay and energy consumption conditions of the tasks processed in the local task queue and unloaded to the edge server corresponding to the execution of the queue; expressing the complete task unloading problem under the multi-user multi-MEC server model in each time slice; and finally, a complete calculation unloading scheme based on near-end strategy optimization is utilized to solve the dynamic target optimization problem of the multi-user multi-MEC server system, so that higher long-term total benefit of the system is realized, the task execution time delay and the expenditure in the aspect of energy consumption of mobile equipment are reduced, and the user experience quality is improved.
The invention provides a multi-server complete computation unloading strategy in a mobile edge environment, which is mainly applied to the technical field of task unloading and resource allocation problems of mobile edge computation and also mainly solves the problem that tasks generated by each mobile user cannot be divided in an application scene of a multi-user multi-MEC server. Aiming at the scene and the task type, a dynamic complete task unloading and resource allocation algorithm based on near-end strategy optimization is designed, detailed definitions of three key elements of a state, an action and a reward function in a Markov decision process are given, and finally a complete calculation unloading model taking minimization of total delay of all task calculation and total energy consumption of terminal equipment as optimization targets is realized.
Compared with the prior art, the invention also has the following advantages:
(1) Under the complete unloading condition of multiple mobile devices and multiple MEC servers, the PPO algorithm in the deep reinforcement learning theory is applied to the calculation unloading problem, and a complete calculation unloading scheme based on PPO is provided.
(2) The proposed scheme allows the execution of tasks to be completed on the mobile device and any MEC server, and can dynamically make reasonable task offloading decisions and allocate bandwidth and computational resources for each user at each time slice, thereby minimizing the total overhead of time delay and energy consumption weighting.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings needed to be used in the embodiments of the present invention will be briefly described below, and it is obvious that the drawings described below are only some embodiments of the present invention, and it is obvious for those skilled in the art that other drawings can be obtained according to the drawings without creative efforts.
Fig. 1 is a flowchart of a multi-server complete computation offload method in a mobile edge environment according to an embodiment of the present invention.
Fig. 2 is a schematic diagram of a multi-server complete computation offload method in a mobile edge environment according to an embodiment of the present invention.
FIG. 3 is a block diagram of a multi-server complete computing offload system in a mobile edge environment according to an embodiment of the present invention;
in the figure: 1. an application scene construction module; 2. a network communication module; 3. a task calculation module; 4. a problem planning module; 5. and an algorithm optimization module.
Fig. 4 is a schematic structural diagram of a multi-server complete computing offload system in a mobile edge environment according to an embodiment of the present invention.
Fig. 5 is a schematic diagram illustrating an influence of the computing power of the MEC server on the total benefit of the system in the simulation experiment node according to the embodiment of the present invention.
Fig. 6 is a schematic diagram illustrating an influence of the number of terminal devices on the total task delay in a simulation experiment result provided by the embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is further described in detail with reference to the following embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.
Aiming at the problems of intensive real-time computing task, weak computing power of a mobile terminal, low running efficiency of a core network and the like in the technical field of the current Internet of things and 5G networks, the invention provides a multi-server complete computing unloading method and a system in a mobile edge environment for application scenes such as intelligent video monitoring, intelligent supply chains, internet of vehicles, industrial Internet and the like, and the invention is described in detail by combining the attached drawings.
As shown in fig. 1, the method for unloading complete computation by multiple servers in a mobile edge environment according to the embodiment of the present invention includes the following steps:
s101, constructing an application scene with a multi-user terminal, a single base station and a multi-MEC server, and realizing unloading decision of tasks of mobile terminal equipment;
s102, calculating a data transmission rate according to the Shannon-Hartley law, constructing a wireless network communication model, and realizing a network bandwidth resource allocation decision of all mobile terminals;
s103, constructing a task calculation model, realizing analysis of time delay and energy consumption under two task processing conditions, and providing basic content for system problem planning;
s104, expressing the complete task unloading problem under the multi-user multi-MEC server model in each time slice, realizing the problem planning of the multi-user multi-server MEC system, and providing an optimization target and related constraints for the unloading scheme;
and S105, providing a complete calculation unloading scheme based on near-end strategy optimization, solving the dynamic target optimization problem of the multi-user multi-MEC server system, and realizing the target of minimizing task delay and energy consumption.
A schematic diagram of a multi-server complete computation offload method in a mobile edge environment according to an embodiment of the present invention is shown in fig. 2.
As shown in fig. 3, the system for unloading multi-server complete computing in a mobile edge environment according to an embodiment of the present invention includes:
the application scene construction module 1 is used for constructing an application scene with a multi-user terminal, a single base station and a multi-MEC server, and realizing the unloading decision of the tasks of the mobile terminal equipment;
the network communication module 2 is used for calculating a data transmission rate according to the Shannon-Hartley law, constructing a wireless network communication model and realizing a network bandwidth resource allocation decision of all the mobile terminals;
the task computing module 3 is used for constructing a task computing model, realizing the analysis of time delay and energy consumption under the condition of processing two tasks and providing basic content for the planning of system problems;
the problem planning module 4 is used for expressing the complete task unloading problem under the multi-user multi-server MEC server model in each time slice, realizing the problem planning of the multi-user multi-server MEC system and providing an optimization target and related constraints for the unloading scheme;
and the algorithm optimization module 5 is used for providing a complete calculation unloading scheme based on near-end strategy optimization, solving the dynamic target optimization problem of the multi-user multi-MEC server system and realizing the target of minimizing task delay and energy consumption.
The structural schematic diagram of the multi-server complete computing offload system in the mobile edge environment provided by the embodiment of the present invention is shown in fig. 4.
The technical solution of the present invention is further described with reference to the following specific examples.
As shown in fig. 2, the specific steps of the multi-server complete computation offload policy in the mobile edge environment provided by the embodiment of the present invention are as follows:
1) And (5) application scene description. And constructing an application scene with a multi-user terminal, a single base station and a multi-MEC server, and unloading a new task which arrives at the mobile terminal device at a certain time slice.
2) And designing a network communication model. And calculating a data transmission rate according to the Shannon-Hartley law, constructing a wireless network communication model transmitted by a plurality of user equipment on orthogonal channels, and making network bandwidth resource allocation decisions for all mobile terminals on a time slice.
3) And constructing a task computing model. The task computing model is divided into a complete local model and an MEC server unloading computing model, and the time delay and the energy consumption condition of the execution of the corresponding queue when the task is processed in the local task queue and unloaded to the edge server are discussed.
4) Problem planning for multi-user multi-server MEC system. The complete task unloading problem under the multi-user multi-MEC server model in each time slice is expressed, and the aim is to minimize the time delay and the energy consumption cost of all tasks under the multi-user multi-server MEC system.
5) And minimizing the weighted minimum sum of the task processing delay and the energy consumption of the mobile equipment in all time slices. The dynamic target optimization problem of the multi-user multi-MEC server system is solved by using a complete computation unloading scheme based on near-end strategy optimization.
The step 1) specifically comprises the following steps:
1.1 Build MEC system application scenarios
Under the system, a multi-user terminal and a multi-MEC server are expressed as follows:
U＝{1,2,...}，C＝{1,2,...}；
each terminal u generates an undetachable resource-intensive task T u ＝{q u ,X u ,δ u U is equal to U, wherein q is equal to u Indicating that terminal u will be tasked with T u Uploading the required input data volume to the selected MEC server from the local; x u Representing the workload, namely the required CPU calculation period number; delta. For the preparation of a coating u Representing a task T u A cut-off time;
1.2 Task offload decisions
The MEC system model divides the total time into time slices with all time lengths, and a new task T is arranged on each time slice T terminal u u To arrive, it needs to be right for task T u Carrying out unloading decision; wherein, the unloading situation comprises two types, one is executed locally, and the other is unloaded to the MEC server for processing, and a binary variable x is defined u E {0,1} is used to denote both cases, where x u =0 represents task T u The calculation is carried out locally by utilizing a local CPU; x is the number of u =1 represents task T u Unloading to an MEC server for processing;
import offload decision x = [ x ] 1 ,x 2 ,...,x |U| ]And y = [ y = 1 ,y 2 ,...,y |U| ]If task T u Offloaded, defining a vector for each mobile user u:
y u ＝[y u,1 ,y u,2 ,...,y u,c ,...,y u,|C| ]；
wherein, y u,c ∈{0,1}，y u,c =1 denotes selecting the task T u Off-loading to edge server c; representing a task T u Not offloaded to edge server c; task T u Can and can only be offloaded to one edge server, i.e. the decision variable y u,c The following constraints are satisfied:
1.3 Set and update task buffer queues:
each user terminal u is provided with a local waiting task buffer queue I which is calculated in a first-in first-out FIFO sequence u If user u chooses to execute task T locally u Add it to the calculation queue I u Tail, K u Is queue I u The above amount of tasks to be performed, the dynamic update formula is expressed as:
K u (t+1)＝max{K u (t)+(1-x u (t))X u (t)-φ u (t),0},u∈U；
wherein, K u (t + 1) the amount of tasks to be performed, x, on the queue is calculated for end user u on time slice t +1 u To indicate task T u Decision variables for local processing or edge offload, X u Is to execute task T u The number of CPU cycles required, phi u (t) represents the local computational workload of the mobile device u within a time slice t;
1.4 Maintain and update task computation queues:
each MEC server is responsible for maintaining | U | task computation queues λ c ＝[λ 1,c ,λ 2,c ,...,λ u,c ,...,λ |U|,c ]And each queue lambda u,c Corresponding to a mobile terminal u; if the edge server receives the unloading task T within the time slice T u Put the task at T u Corresponding queue lambda of incoming MEC server u,c In, update the formulaComprises the following steps:
wherein M is u,c (t + 1) represents the amount of tasks to be executed in the task queue maintained by the edge server c for the terminal u in the time slice t +1, y u,c To indicate task T u A decision variable whether to be offloaded to server c,
then it is the calculation queue λ within time slice t u,c The amount of tasks performed.
Step 2) above, network communication model is established
In the network communication model, assuming that the total available bandwidth of a wireless channel in the wireless network communication model in which a plurality of user equipment transmit on orthogonal channels is B, the uplink data transmission rate between the tth time slice mobile user u and the MEC server c is represented by r u,c It is shown that the network bandwidth resource allocation decision made for all mobile terminals over time slice t is represented as:
b＝[b 1 ,b 2 ,...,b u ,...,b |U| ]；
wherein, b u Ratio of bandwidth resources, P, allocated to the u mobile terminal by the base station for the t time slice u For the data transmission power of the mobile terminal u, N 0 Is the Gaussian white noise channel power, h u,c Representing the channel gain between the t time slice user equipment u and the edge server c, the uplink data transmission rate r between the t time slice mobile user u and the MEC server c u,c The calculation formula is as follows:
proportion b of network uplink bandwidth resources allocated to each user terminal in time slice t u The following constraints are satisfied:
in the task unloading model of the step 3), the user u carries out the task T u Is made of a variable x u And y u,c Decide together, so discuss task T separately u In the local task queue I u Processing and offloading to a corresponding queue λ on edge server c u,c The time delay and energy consumption of execution are divided into the following steps according to two situations:
3.1 Build a complete local computation model
When unloading variable x u =0, i.e. selecting task T u Adding into a local waiting processed task buffer queue I u Time of day, it calculates the time delay locally
Can be expressed as:
wherein, X u Representing a processing task T u The number of cycles required for the CPU to count,
is the local CPU computation power of end user u. In addition, task T u Waiting for local buffer queue I u The above tasks can be executed after all the tasks are processed, and the waiting time of the process can be calculated as:
thus task T u The local total service delay of (a) is expressed as follows:
task T u Selecting energy consumption when executing locally
By computing task T according to a dynamic voltage scaling technique u Working load and local CPU working frequency of
It can be calculated as:
where κ is an effective switched capacitance parameter that depends on the chip structure.
3.2 Build MEC Server offload computation model
When unloading variable x u 1 and y u,c When =1, i.e. task T u Is decided to be offloaded to the c-th MEC server. Requiring the input of data D by a base station for a task u Is forwarded. Meanwhile, in the edge unloading model, a reasonable calculation resource proportion distribution decision F needs to be considered to be made for | U | task calculation queues on the server c c ＝[F 1,c ,F 2,c ,...,F u,c ,...,F |U|,c ]And for each edge server c the following constraints should be satisfied:
the terminal device u sends the task T u Time delay of unloading to c MEC server through base station
And energy consumption
Respectively as follows:
wherein, y u,c Is to indicate task T u Decision variable, q, whether to offload to Server c u Representing a task T u Size of the uploaded data amount r u,c For the uplink transmission rate between device u and server c, obtained according to the above network communication model, P u Is the transmit power of the u-th user terminal.
Given task queue lambda u,c Is calculated by the computing resource ratio allocation of F u,c Then task T u Execution time on edge Server c
Comprises the following steps:
wherein f is c MEC Representing the CPU computing power of the MEC server c.
Task T u Corresponding task queue lambda of the c-th MEC server u,c Wait time of
Comprises the following steps:
task T on end user u under edge offload u Total service delay of
Can be calculated as:
the total energy consumption of the end user u is the task T unloaded to the edge server c u The transmission energy consumption of (a), namely:
the multi-user multi-server task computing model established in the step 4) is as follows:
wherein the content of the first and second substances,
and
respectively representing tasks T generated by the u th mobile device of the current time slice u The execution latency of the full local computation and full offload to the server side,
and
respectively represent tasks T u The transmission energy consumption of the mobile device u under the conditions of local execution energy consumption and edge unloading is introduced, and a parameter mu is introduced 1 And mu 2 To respectively express the weight of time and energy consumption cost and satisfy mu 1 +μ 2 ＝1；
Wherein C1 in the constraint condition represents a task T u Whether to unload to the server side for execution; c2 denotes indicating task T u A variable whether it is offloaded to server c; c3 and C4 respectively indicate the bandwidth resource ratio of the base station to the terminal user, namely the bandwidth resource ratio of the base station to the terminal user and the bandwidth resource ratio of the MEC server to the terminalThe proportion of computing resources of the end user is not more than 1; c5 indicates that the processing latency of the task does not exceed a given maximum deadline.
The step 5) of completely calculating the unloading strategy based on the PPO comprises the following steps:
5.1 Markov decision plan description
On the basis of the reinforcement learning framework, the state, the action and the reward function of the multi-user multi-server MEC system in the time slice t are respectively defined in detail as follows:
5.1.1 State): the state S (t) ∈ S of the MDP on the time slice t consists of six parts, respectively the data volume size per mobile device task D (t) = [ D) = 1 (t),D 2 (t),...,D U (t)]The number of computing resources required to complete a task X (t) = [ X = [) 1 (t),X 2 (t),...,X |U| (t)]The amount of tasks to be executed on each mobile equipment calculation queue K (t) = [ K = [) 1 (t),K 2 (t),...,K |U| (t)]The method comprises the following steps that an MEC server task quantity matrix M (t) to be executed and an uplink transmission rate matrix r (t), wherein the sizes of the M (t) matrix and the r (t) matrix are | U | × | C |, and elements in the matrix are M respectively u,c And r u,c And representing the amount of tasks to be executed by the task queue maintained by the MEC server c for the mobile device u and the uplink transmission rate between the device u and the MEC server c, the state s (t) is described as s (t) = [ D (t), X (t), K (t), M (t), r (t)]。
5.1.2 Action): the action space A of the MDP includes three parts of decision making, respectively task T u Made offload decision x (t) = [ x = [) 1 (t),x 2 (t),...,x |U| (t)]And y (t) = [ y = u,c (t)]U ∈ U, C ∈ C, the network bandwidth resource proportion allocation decision b (t) = [ b ] made by each mobile user 1 (t),b 2 (t),...,b |U| (t)]And the computation resource proportion allocation decision F (t) = [ F) made by the edge server for all mobile users u,c (t)]U is U, C is C, wherein F is u,c (t) represents the proportion of computing resources allocated by the MEC server c to the user u over the time slice t, then the action a (t) over the time slice t is defined as a (t) = [ x (t), y (t), b (t), F (t)]。
5.1.3 ) rewardsThe function reward: defining the reward function R (t) as a weighted sum of the calculated delay and energy consumption of all mobile device generated tasks per time slice t, expressed as the target value given by the formula
The goal of the algorithm is to maximize long-term rewards
5.2 Optimization with PPO algorithm):
in order to solve the defects of the strategy gradient algorithm, the near-end strategy optimization algorithm is improved in both stability and convergence, and the calculation unloading strategy of the MEC system provided by the invention is optimized in a more stable and more effective sampling mode.
5.2.1 PPO algorithm optimization goal:
the goal of the PPO algorithm is to maximize the objective function L constrained by the policy update size:
wherein the content of the first and second substances,
the probability ratio between the new strategy and the old strategy is obtained; meanwhile, the PPO algorithm introduces a clip function in the target function to control the value of γ (θ), which is expressed as follows:
where ε is the over-parameter, the clip function is used to limit the probability ratio γ (θ) to the range of [1- ε,1+ ε ]. The main purpose is to effectively adjust the size of the policy update in a way that selects the minimum of clipped and unclipped targets.
5.2.2 PPO-based full computation offload flow:
step1: input task request set T 1 ,T 2 ,...,T |U| B, total bandwidth of wireless channel, computing power f of each MEC server c MEC ；
Step2: calculating a dominance estimate
Using a combination of pi θ Updating old policy π old ；
Step3: repeating the step (2) for N times, wherein N is the number of state conversion data in each epsilon;
step4: optimizing an objective function L by using w epochs and a small batch size sigma < = N eta;
step5: updating network parameter theta with theta according to gradient descent method old ；
Step6: repeating the step (2) to the step (5) for G times, wherein G is the number of epsilon;
step7: and (3) outputting: task unloading decision x and y, wireless bandwidth resource allocation decision b, and MEC server computing resource allocation decision F.
1. Conditions of the experiment
The method utilizes python 3.6 to carry out simulation experiment on the complete calculation unloading algorithm based on the near-end strategy optimization, divides the system time into 1000 time slices, and assumes that the data size of the mobile device generating the task, the calculation resource amount required by the task and the task completion deadline all obey uniform distribution within a certain range on each time slice.
2. Content of the experiment
The strategy provided by the invention is different from the strategy in the existing literature, the performance of the PPO-based complete calculation unloading algorithm is evaluated, a simulation comparison experiment is carried out, and the algorithm is respectively compared with four reference algorithms of local execution, polling scheduling, random unloading and DQN (Deep Q Network, DQN) -based complete calculation unloading, and the method specifically comprises the following steps:
(1) Local execution Algorithm (AL): the tasks that the mobile device arrives at are all performed locally at the device for each time slice.
(2) Round-Robin algorithm (RR): and distributing the tasks submitted by each mobile user to a plurality of MEC servers in turn for processing, and averagely distributing the computing resources of each MEC server to the queue maintained for each end user.
(3) Random Offloading algorithm (RO): tasks submitted by each mobile user are randomly allocated to the MEC servers for processing and the computing resources of each MEC server are evenly allocated to the queues maintained for each end user.
(4) DQN-based complete computation offload algorithm: the allocation of bandwidth resources to the proportional decision b to be made for each end user u u And computing resource allocation proportion decision F u,c All set level =6, which can be respectively expressed as
And
3. results of the experiment
As shown in fig. 5, computing power f for MEC server c MEC The change of the total benefit of the calculation unloading strategy based on PPO and the four comparison algorithm systems provided by the research institute is shown as f c MEC The results are shown in Table 2, with the respective settings being 3, 4, 5, 6, 7 Gcycles/s. Experimental results show that the PPO algorithm provided by the invention has higher total system benefits than other four comparison algorithms, and for the other four algorithms, the total system benefits are improved along with the increase of the computing capacity of the MEC server.
TABLE 2 Effect of MEC Server computing capacity on overall System benefit
As shown in fig. 6, a simulation experiment was performed on five algorithms according to the change of the number of the terminal devices, and the influence of the change of the number of the terminal devices on the total benefit of the system was studied. Table 3 shows the variation of the total processing delay of five algorithm tasks as the number of mobile devices increases. Experimental results show that the PPO-based complete calculation unloading algorithm provided by the invention has optimal performance and slow growth speed in the aspect of total task delay.
TABLE 3 influence of the number of terminal devices on the total time delay of the task
The method carries out a large number of simulation experiments aiming at different parameters such as the calculation capacity of the server, the number of terminal equipment, the number of MEC servers and the like, and compares the simulation experiments with a DQN-based complete calculation unloading algorithm and three baseline strategies of AL local execution, RR polling scheduling and RO random unloading. Simulation experiment results prove that compared with other four algorithms, the PPO-based complete calculation unloading algorithm provided by the invention can better optimize the total benefit of the system and reduce the task execution delay and the overhead of the energy consumption of the mobile equipment.
In the above embodiments, the implementation may be wholly or partially realized by software, hardware, firmware, or any combination thereof. When used in whole or in part, can be implemented in a computer program product that includes one or more computer instructions. When the computer program instructions are loaded or executed on a computer, the procedures or functions according to the embodiments of the present invention are wholly or partially generated. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable device. The computer instructions may be stored in a computer readable storage medium or transmitted from one computer readable storage medium to another, for example, the computer instructions may be transmitted from one website site, computer, server, or data center to another website site, computer, server, or data center via wire (e.g., coaxial cable, fiber optic, digital Subscriber Line (DSL), or wireless (e.g., infrared, wireless, microwave, etc.)). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device, such as a server, a data center, etc., that includes one or more of the available media. The usable medium may be a magnetic medium (e.g., floppy Disk, hard Disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid State Disk (SSD)), among others.
The above description is only for the purpose of illustrating the embodiments of the present invention, and the scope of the present invention should not be limited thereto, and any modifications, equivalents and improvements made by those skilled in the art within the technical scope of the present invention as disclosed in the present invention should be covered by the scope of the present invention.
Claims (5)
1. A multi-server complete computation unloading method in a mobile edge environment is characterized by comprising the following steps:
constructing an application scene with a plurality of mobile terminals, a single base station and a plurality of MEC servers, and realizing the unloading decision of the tasks of the mobile terminal equipment;
step two, calculating a data transmission rate according to a Shannon-Hartley law, constructing a wireless network communication model, and realizing a network bandwidth resource allocation decision of all mobile terminals;
step three, constructing a task calculation model, realizing the analysis of time delay and energy consumption under the condition of processing two tasks, and providing basic content for the planning of system problems;
expressing the complete task unloading problem under the multi-mobile-terminal multi-MEC server model in each time slice, realizing the problem planning of the multi-mobile-terminal multi-server MEC system, and providing an optimization target and related constraints for the unloading scheme;
step five, a complete calculation unloading scheme based on near-end strategy optimization is provided, the dynamic target optimization problem of a multi-mobile-terminal multi-MEC server system is solved, and the target of minimizing task delay and energy consumption is realized;
in the first step, an application scene with a plurality of mobile terminals, a single base station and a plurality of MEC servers deployed is constructed to realize the unloading decision of the tasks of the mobile terminal equipment, and the method comprises the following steps:
(1) Constructing MEC system application scenarios
Under the system, a plurality of mobile terminals and a plurality of MEC servers are expressed as follows:
U＝{1,2,...}，C＝{1,2,...}；
each terminal u generates an undetachable resource-intensive task T u ＝{q u ,X u ,δ u U is equal to U, wherein q is equal to u Indicating that terminal u will be tasked with T u Uploading the required input data volume to the selected MEC server from the local; x u Representing the workload, namely the required CPU calculation period number; delta u Representing a task T u A cut-off time;
(2) Task offload decision
The MEC system model divides the total time into time slices with all time lengths, and a new task T is arranged on each time slice T terminal u u To arrive, it needs to be right for task T u Carrying out unloading decision; the unloading situation comprises two situations, wherein one situation is executed locally, and the other situation is unloaded to the MEC server for processing, and a binary variable x is defined u E {0,1} is used to represent both cases, where x u =0 represents task T u The calculation is carried out locally by utilizing a local CPU; x is the number of u =1 represents task T u Unloading to an MEC server for processing;
import offload decision x = [ x ] 1 ,x 2 ,...,x |U| ]And y = [ y = 1 ,y 2 ,...,y |U| ]If task T u Is offloaded, defining a vector for each mobile terminal u:
y u ＝[y u,1 ,y u,2 ,...,y u,c ,...,y u,|C| ]；
wherein, y u,c ∈{0,1}，y u,c =1 denotes selecting the task T u Off-loading to edge server c; representing a task T u Not offloaded to edge server c; task T u Can and can only be offloaded to one edge server, i.e. the decision variable y u,c The following constraints are satisfied:
(3) Setting and updating a task buffer queue:
each mobile terminal u is provided with a local waiting processed task buffer queue I which is calculated by a first-in first-out FIFO sequence u If the mobile terminal u selects to execute the task T locally u Add it to the calculation queue I u Tail, K u Is queue I u The above amount of tasks to be performed, the dynamic update formula is expressed as:
K u (t+1)＝max{K u (t)+(1-x u (t))X u (t)-φ u (t),0},u∈U；
wherein, K u (t + 1) calculating the amount of tasks to be executed on the queue, x, for the mobile terminal u on the time slice t +1 u To indicate task T u Decision variables for local processing or edge offload, X u Is to execute task T u The number of CPU cycles required, phi u (t) represents the local computational workload of the mobile device u within a time slice t;
(4) Maintaining and updating task computation queues:
each MEC server is responsible for maintaining | U | task computation queues λ c ＝[λ 1,c ,λ 2,c ,...,λ u,c ,...,λ |U|,c ]And each queue λ u,c Corresponding to a mobile terminal u; if the edge server receives the unloading task T within the time slice T u Put the task T u Corresponding queue lambda of incoming MEC server u,c In, the update formula is:
wherein M is u,c (t + 1) represents the amount of tasks to be executed in the task queue maintained by the edge server c for the terminal u in the time slice t +1, y u,c To indicate task T u The decision variable whether or not to be offloaded to server c,
then it is the calculation queue λ within time slice t u,c The amount of tasks performed;
in the second step, the data transmission rate is calculated according to the Shannon-Hartley law, a wireless network communication model is constructed, and the decision of network bandwidth resource allocation of all mobile terminals is realized, and the decision comprises the following steps:
in the network communication model, assuming that the total available bandwidth of a wireless channel in the wireless network communication model in which a plurality of mobile terminals transmit on orthogonal channels is B, the uplink data transmission rate between the tth time slice mobile terminal u and the MEC server c is represented by r u,c It is shown that the network bandwidth resource allocation decision made for all mobile terminals over time slice t is represented as:
b＝[b 1 ,b 2 ,...,b u ,...,b |U| ]；
wherein, b u Ratio of bandwidth resources, P, allocated to the u mobile terminal by the base station for the t time slice u For the data transmission power of the mobile terminal u, N 0 Is the Gaussian white noise channel power, h u,c Represents the channel gain between the t time slice mobile terminal u and the edge server c, the uplink data transmission rate r between the t time slice mobile terminal u and the MEC server c u,c The calculation formula is as follows:
proportion b of network uplink bandwidth resources allocated to each mobile terminal in time slice t u The following constraints are satisfied:
in step three, the task calculation model is constructed to realize the analysis of time delay and energy consumption under the condition of processing two tasks, and basic contents are provided for the planning of system problems, and the method comprises the following steps:
in the task unloading model, a task T is arranged on a mobile terminal u u Is made of a variable x u And y u,c Decide together, so analyze task T separately u In the local task queue I u Processing and offloading to a corresponding queue λ on edge server c u,c The time delay and energy consumption situation of execution are divided into the following steps according to two situations:
(1) Building a complete local computation model
When unloading variable x u =0, i.e. selecting task T u Adding into a local waiting processed task buffer queue I u Time of day, it calculates the time delay locally
Expressed as:
wherein, X u Representing a processing task T u The number of cycles required for the CPU to calculate,
is the local CPU computing power of the mobile terminal u; in addition, task T u Waiting for local buffer queue I u The above tasks can be executed after all the tasks are processed, and the waiting time of the process is calculated as:
thus task T u The local total service delay of (a) is expressed as follows:
task T u Selecting energy consumption when executing locally
By computing task T according to a dynamic voltage scaling technique u Working load and local CPU working frequency of
It can be calculated as:
wherein κ is an effective switched capacitance parameter dependent on the chip structure;
(2) Constructing MEC server offload computation model
When unloading variable x u =1 and y u,c When =1, i.e. task T u Is decided to be unloaded to the c MEC server, and the terminal device u unloads the task T u Time delay offloaded to the c-th MEC server by the base station
And energy consumption
Respectively as follows:
wherein, y u,c Is to indicate task T u Decision variable, q, whether to offload to Server c u Representing a task T u Size of the uploaded data amount r u,c To be in accordance with the above network communication modelThe obtained uplink transmission rate, P, between the device u and the server c u Is the transmission power of the u-th mobile terminal;
given task queue lambda u,c Is calculated by the computing resource ratio allocation of F u,c Then task T u Execution time on edge server c
Comprises the following steps:
wherein f is c MEC Represents the CPU computing power of the MEC server c;
task T u Corresponding task queue lambda of the c-th MEC server u,c Waiting time of
Comprises the following steps:
task T on mobile terminal u under condition of edge unloading u Total service delay of
The calculation is as follows:
the total energy consumption of the mobile terminal u is the task T unloaded to the edge server c u The transmission energy consumption of (a), namely:
in the fourth step, the complete task unloading problem under the multi-mobile-terminal multi-MEC server model in each time slice is expressed, the problem planning of the multi-mobile-terminal multi-server MEC system is realized, and an optimization target and related constraints are provided for the unloading scheme, wherein the method comprises the following steps:
the multi-mobile terminal and multi-server task calculation model is established as follows:
wherein the content of the first and second substances,
and
respectively representing tasks T generated by the u th mobile device of the current time slice u Full local computation and full offload to server sideThe execution time delay of (2) is,
and
respectively represent tasks T u The transmission energy consumption of the mobile device u under the conditions of local execution energy consumption and edge unloading is introduced, and a parameter mu is introduced 1 And mu 2 To respectively express the weight of time and energy consumption cost and satisfy mu 1 +μ 2 ＝1；
Wherein, C1 in the constraint condition represents a task T u Whether to unload to the server side for execution; c2 denotes indicating task T u A variable whether it is offloaded to server c; c3 and C4 respectively indicate that the bandwidth resource ratio allocated to the mobile terminal by the base station refers to the bandwidth resource ratio allocated to the mobile terminal by the base station and the calculation resource ratio allocated to the mobile terminal by the MEC server, and both are not more than 1; c5 represents that the processing time delay of the task does not exceed a given maximum cut-off time;
in the fifth step, the proposed complete computation unloading scheme based on the near-end strategy optimization solves the dynamic target optimization problem of the multi-mobile terminal multi-MEC server system, and achieves the target of minimizing task delay and energy consumption, and comprises the following steps:
(1) Markov decision plan description
On the basis of the reinforcement learning framework, the state, the action and the reward function of the multi-mobile-terminal multi-server MEC system in the time slice t are respectively defined in detail as follows:
1) State: the state S (t) ∈ S of the MDP on a time slice t consists of six parts, respectively the data size per mobile device task D (t) = [ D = [ D ] ] 1 (t),D 2 (t),...,D |U| (t)]The number of computing resources required to complete a task X (t) = [ X = [) 1 (t),X 2 (t),...,X |U| (t)]The amount of tasks to be executed on each mobile equipment calculation queue K (t) = [ K = [) 1 (t),K 2 (t),…,K |U| (t)]An MEC server to-be-executed task quantity matrix M (t) and an uplink transmission rate matrix r (t) which areThe size of the matrix of M (t) and r (t) is | U | × | C |, and the elements in the matrix are M respectively u,c And r u,c Representing the amount of tasks to be performed by the task queue maintained by the MEC server c for the mobile device u and the upstream transmission rate between the device u and the MEC server c, the state s (t) is described as s (t) = [ D (t), X (t), K (t), M (t), r (t)]；
2) Action: the action space A of the MDP includes three parts of decision making, respectively task T u Made offload decision x (t) = [ x = [) 1 (t),x 2 (t),…,x |U| (t)]And y (t) = [ y = u,c (t)]U ∈ U, C ∈ C, the network bandwidth resource proportion allocation decision b (t) = [ b ] made by each mobile terminal 1 (t),b 2 (t),…,b |U| (t)]And the computation resource proportion allocation decision F (t) = [ F ] made by the edge server for all mobile terminals u,c (t)]U ∈ U, C ∈ C, wherein F ∈ C u,c (t) represents the proportion of computing resources allocated by the MEC server c to the mobile terminal u over the time slice t, then the action a (t) over the time slice t is defined as a (t) = [ x (t), y (t), b (t), F (t)]；
3) Reward function reward: defining the reward function R (t) as a weighted sum of the calculated delay and energy consumption of all mobile device generated tasks per time slice t, expressed as the target value given by the formula
The goal of the algorithm is to maximize long-term rewards
(2) Optimization with near-end policy optimization PPO algorithm
1) PPO algorithm target
The goal of the PPO algorithm is to maximize the objective function L constrained by the policy update size:
wherein, the first and the second end of the pipe are connected with each other,
the probability ratio between the new strategy and the old strategy is obtained; meanwhile, the PPO algorithm introduces a clip function in the target function to control the value of γ (θ), which is expressed as follows:
wherein epsilon is a hyper-parameter, the clip function is used for limiting the probability ratio gamma (theta) within the range of [ 1-epsilon, 1+ epsilon ], and the size of strategy updating is effectively adjusted in a mode of selecting the minimum value of the clipped target and the unclipped target;
2) PPO-based complete calculation unloading process
(1) Input task request set T 1 ,T 2 ,...,T |U| }, total bandwidth of wireless channel B, computation power per MEC server f c MEC ；
(2) Calculating a dominance estimate
Using a combination of pi θ Updating old policy π old ；
(3) Repeating the step (2) for N times, wherein N is the number of state conversion data in each epsilon;
(4) optimizing an objective function L by using w epochs and a small batch size sigma < = N eta;
(5) updating the network parameter theta by theta according to the gradient descent method old ；
(6) Repeating the step (2) to the step (5) for G times, wherein G is the number of epsilon;
(7) and (3) outputting: task unloading decision x and y, wireless bandwidth resource allocation decision b, and MEC server calculation resource allocation decision F.
2. A multi-server complete computing offload system in a mobile edge environment for implementing the multi-server complete computing offload method in the mobile edge environment of claim 1, wherein the multi-server complete computing offload system in the mobile edge environment comprises:
the application scene construction module is used for constructing an application scene with a plurality of mobile terminals, a single base station and a plurality of MEC servers, and realizing the unloading decision of the tasks of the mobile terminal equipment;
the network communication module is used for calculating the data transmission rate according to the Shannon-Hartley law, constructing a wireless network communication model and realizing the network bandwidth resource allocation decision of all the mobile terminals;
the task computing module is used for constructing a task computing model, realizing the analysis of time delay and energy consumption under the condition of processing two tasks and providing basic content for the planning of system problems;
the problem planning module is used for expressing the complete task unloading problem under the multi-mobile-terminal multi-MEC server model in each time slice, realizing the problem planning of the multi-mobile-terminal multi-server MEC system and providing an optimization target and related constraints for the unloading scheme;
and the algorithm optimization module is used for providing a complete calculation unloading scheme based on near-end strategy optimization, solving the dynamic target optimization problem of the multi-mobile-terminal multi-MEC server system and realizing the target of minimizing task delay and energy consumption.
3. A computer arrangement comprising a memory and a processor, the memory storing a computer program that, when executed by the processor, causes the processor to perform the method of multi-server full computation offload in a mobile edge environment of claim 1.
4. A computer-readable storage medium storing a computer program which, when executed by a processor, causes the processor to execute the multi-server complete computing offload method in a mobile edge environment of claim 1.
5. An information data processing terminal characterized by being used for implementing a multi-server complete computing offload system in a mobile edge environment according to claim 2.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111061668.3A CN113950103B (en) | 2021-09-10 | 2021-09-10 | Multi-server complete computing unloading method and system under mobile edge environment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111061668.3A CN113950103B (en) | 2021-09-10 | 2021-09-10 | Multi-server complete computing unloading method and system under mobile edge environment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113950103A CN113950103A (en) | 2022-01-18 |
| CN113950103B true CN113950103B (en) | 2022-11-04 |
Family
ID=79328011
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111061668.3A Active CN113950103B (en) | 2021-09-10 | 2021-09-10 | Multi-server complete computing unloading method and system under mobile edge environment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113950103B (en) |
Families Citing this family (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114461299B (en) * | 2022-01-26 | 2023-06-06 | 中国联合网络通信集团有限公司 | Unloading decision determining method and device, electronic equipment and storage medium |
| CN114513838A (en) * | 2022-02-28 | 2022-05-17 | 中国农业银行股份有限公司 | Moving edge calculation method, frame, and medium |
| CN114615705B (en) * | 2022-03-11 | 2022-12-20 | 广东技术师范大学 | Single-user resource allocation strategy method based on 5G network |
| CN114640675B (en) * | 2022-03-21 | 2024-02-09 | 中国联合网络通信集团有限公司 | Unloading strategy determining method and device, electronic equipment and storage medium |
| CN114760311B (en) * | 2022-04-22 | 2023-09-05 | 南京邮电大学 | Optimized service caching and computing unloading method for mobile edge network system |
| CN115002801B (en) * | 2022-04-27 | 2024-04-16 | 燕山大学 | Edge computing resource dynamic unloading method and device based on passive relay collaboration |
| CN114884945B (en) * | 2022-04-28 | 2023-08-11 | 广东电网有限责任公司 | Data transmission method, cloud server, device, system and storage medium |
| CN115002113B (en) * | 2022-05-26 | 2023-08-01 | 南京邮电大学 | Mobile base station edge computing power resource scheduling method, system and electronic equipment |
| CN115022189B (en) * | 2022-05-31 | 2024-03-26 | 武汉大学 | Edge user allocation model construction method, device, equipment and readable storage medium |
| CN115051998B (en) * | 2022-06-09 | 2023-06-20 | 电子科技大学 | Adaptive edge computing offloading method, apparatus and computer-readable storage medium |
| CN115225496A (en) * | 2022-06-28 | 2022-10-21 | 重庆锦禹云能源科技有限公司 | Mobile sensing service unloading fault-tolerant method based on edge computing environment |
| CN115190126B (en) * | 2022-07-01 | 2023-08-18 | 北京理工大学长三角研究院(嘉兴) | Mobile edge computing system for coordinating computation and transmission and optimal unloading method |
| CN115225675A (en) * | 2022-07-18 | 2022-10-21 | 国网信息通信产业集团有限公司 | Charging station intelligent operation and maintenance system based on edge calculation |
| CN115396953A (en) * | 2022-08-02 | 2022-11-25 | 黑龙江八一农垦大学 | Calculation unloading method based on improved particle swarm optimization algorithm in mobile edge calculation |
| CN115396955A (en) * | 2022-08-24 | 2022-11-25 | 广西电网有限责任公司 | Resource allocation method and device based on deep reinforcement learning algorithm |
| CN115426033B (en) * | 2022-11-03 | 2023-03-24 | 中国电子科技集团公司第二十八研究所 | Community queue control method based on Sarsa algorithm |
| CN116112976B (en) * | 2022-12-20 | 2024-05-03 | 暨南大学 | Equipment calculation migration method, device, equipment and storage medium |
| CN115629865B (en) * | 2022-12-20 | 2023-04-18 | 合肥综合性国家科学中心人工智能研究院(安徽省人工智能实验室) | Deep learning inference task scheduling method based on edge calculation |
| CN116170846B (en) * | 2023-04-25 | 2023-07-14 | 天津市工业和信息化研究院(天津市节能中心、天津市工业和信息化局教育中心) | Distributed edge computing system and edge computing method for data communication |
| CN116668447B (en) * | 2023-08-01 | 2023-10-20 | 贵州省广播电视信息网络股份有限公司 | Edge computing task unloading method based on improved self-learning weight |
| CN117194057B (en) * | 2023-11-08 | 2024-01-23 | 贵州大学 | Resource scheduling method for optimizing edge energy consumption and load based on reinforcement learning |
| CN117873689A (en) * | 2024-03-11 | 2024-04-12 | 浪潮计算机科技有限公司 | Task allocation method, device, equipment and computer readable storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108920279A (en) * | 2018-07-13 | 2018-11-30 | 哈尔滨工业大学 | A kind of mobile edge calculations task discharging method under multi-user scene |
| CN112860350A (en) * | 2021-03-15 | 2021-05-28 | 广西师范大学 | Task cache-based computation unloading method in edge computation |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2020023115A1 (en) * | 2018-07-27 | 2020-01-30 | Futurewei Technologies, Inc. | Task offloading and routing in mobile edge cloud networks |
| CN109814951B (en) * | 2019-01-22 | 2021-09-28 | 南京邮电大学 | Joint optimization method for task unloading and resource allocation in mobile edge computing network |
| CN110493360B (en) * | 2019-09-26 | 2022-08-19 | 重庆大学 | Mobile edge computing unloading method for reducing system energy consumption under multiple servers |
| CN112381211B (en) * | 2020-11-20 | 2023-04-28 | 西安电子科技大学 | System and method for executing deep neural network based on heterogeneous platform |
| CN113225377B (en) * | 2021-03-30 | 2022-07-15 | 北京中电飞华通信有限公司 | Internet of things edge task unloading method and device |
| CN113242568B (en) * | 2021-06-08 | 2022-05-03 | 重庆邮电大学 | Task unloading and resource allocation method in uncertain network environment |
-
2021
- 2021-09-10 CN CN202111061668.3A patent/CN113950103B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108920279A (en) * | 2018-07-13 | 2018-11-30 | 哈尔滨工业大学 | A kind of mobile edge calculations task discharging method under multi-user scene |
| CN112860350A (en) * | 2021-03-15 | 2021-05-28 | 广西师范大学 | Task cache-based computation unloading method in edge computation |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113950103A (en) | 2022-01-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113950103B (en) | Multi-server complete computing unloading method and system under mobile edge environment | |
| CN113950066B (en) | Single server part calculation unloading method, system and equipment under mobile edge environment | |
| CN109684075B (en) | Method for unloading computing tasks based on edge computing and cloud computing cooperation | |
| CN111586762B (en) | Task unloading and resource allocation joint optimization method based on edge cooperation | |
| CN108920280B (en) | Mobile edge computing task unloading method under single-user scene | |
| CN111930436B (en) | Random task queuing unloading optimization method based on edge calculation | |
| CN112039965B (en) | Multitask unloading method and system in time-sensitive network | |
| WO2023040022A1 (en) | Computing and network collaboration-based distributed computation offloading method in random network | |
| CN113286329B (en) | Communication and computing resource joint optimization method based on mobile edge computing | |
| CN113553165B (en) | Game theory-based mobile edge computing task unloading and resource scheduling method | |
| CN110149401B (en) | Method and system for optimizing edge calculation task | |
| CN114928607B (en) | Collaborative task unloading method for polygonal access edge calculation | |
| CN113254095B (en) | Task unloading, scheduling and load balancing system and method for cloud edge combined platform | |
| CN113992678A (en) | Calculation migration method for offshore MEC load balancing and resource allocation joint optimization | |
| Zhang et al. | Effect: Energy-efficient fog computing framework for real-time video processing | |
| CN116366576A (en) | Method, device, equipment and medium for scheduling computing power network resources | |
| CN114172558B (en) | Task unloading method based on edge calculation and unmanned aerial vehicle cluster cooperation in vehicle network | |
| CN111158893B (en) | Task unloading method, system, equipment and medium applied to fog computing network | |
| CN112817741B (en) | DNN task control method for edge calculation | |
| CN110768827B (en) | Task unloading method based on group intelligent algorithm | |
| CN116302404B (en) | Resource decoupling data center-oriented server non-perception calculation scheduling method | |
| CN111580943A (en) | Task scheduling method oriented to multi-hop unloading in low-delay edge calculation | |
| CN115150893A (en) | MEC task unloading strategy method based on task division and D2D | |
| Mahjoubi et al. | Optimal placement of recurrent service chains on distributed edge-cloud infrastructures | |
| CN114143317A (en) | Cross-cloud-layer mobile edge calculation-oriented multi-priority calculation unloading strategy optimization method |
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 |