A joint optimization method for task offloading and resource allocation in containerized edge computing

Abstract

The present invention discloses a joint optimization method for task offloading and resource allocation of containerized edge computing, which belongs to the field of task offloading and resource allocation in edge computing scenarios. The method includes: establishing a containerized edge computing system model; establishing a joint optimization problem according to the established mathematical model; decomposing a multi-time slot problem into a single time slot problem; and linearizing the nonlinear terms in the original problem by an equivalent transformation method. The present invention adopts the above-mentioned joint optimization method for task offloading and resource allocation of containerized edge computing, which can quickly obtain the joint optimization optimal solution of task offloading and resource allocation in the edge computing environment, and reduce the task processing time of terminal devices in the containerized edge computing system.

Description

Technical Field

[0001] The present invention relates to the technical field of task offloading and resource allocation in edge computing scenarios, and in particular to a joint optimization method for task offloading and resource allocation in containerized edge computing. Background Art

[0002] At present, the two main methods for the joint optimization of task offloading and resource allocation in edge computing environments are alternating iterative optimization and deep reinforcement learning. Among them, the alternating iterative optimization method usually uses the block coordinate descent method to alternately optimize task offloading decisions and service caching decisions when solving coupling problems, but this will cause the algorithm to converge very slowly. And when solving non-convex problems, due to the limitations of block coordinate descent, alternating iterative optimization cannot find the global optimal solution to the problem very well. When solving optimization problems, deep reinforcement learning needs to obtain the system parameters in advance and train a suitable model by building an environment. Although the model obtained through training can solve the problem quickly, due to the long model training time, deep reinforcement learning can often only be used as an offline algorithm, which will lead to serious limitations in practical applications.

[0003] The present invention proposes a joint optimization method for task offloading and resource allocation in containerized edge computing, which is a joint optimization algorithm for task offloading and resource allocation based on linear transformation, and can quickly obtain the joint optimization optimal solution of task offloading and resource allocation in an edge computing environment. Summary of the invention

[0004] The purpose of the present invention is to provide a method for joint optimization of task offloading and resource allocation in containerized edge computing, which can quickly obtain the optimal solution for the joint optimization of task offloading and resource allocation in an edge computing environment.

[0005] To achieve the above object, the present invention provides a method for joint optimization of task offloading and resource allocation of containerized edge computing, comprising the following steps:

[0006] S1. Establish a containerized edge computing system model, including a cloud server layer, an edge server layer, and a terminal device layer, and establish a mathematical model based on characteristic parameters of the cloud server, edge server, terminal device, and container involved in the edge computing environment;

[0007] S2, establish a multi-slot nonlinear mixed integer programming problem based on the mathematical model;

[0008] S3, decompose the multi-slot problem into a single-slot problem;

[0009] S4. Convert the nonlinear mixed integer programming problem into a linear mixed integer programming problem through equivalent transformation.

[0010] Preferably, the specific content of S1 is:

[0011] S11. Using queue modeling, introduce terminal task computing queue and terminal task transmission queue in terminal devices, and introduce edge task computing queue in edge servers. The content is as follows:

[0012] gather Represents a collection of terminal devices, ;gather Represents a collection of edge servers, ;gather Represents a collection of images, ;gather represents the set of time slots of the system, ;

[0013] Terminal task calculation queue:

[0014] ;

[0015] in, Indicates terminal equipment In time slot The tasks generated The amount of tasks, Indicates terminal equipment Mission In time slot The task offloading decision Indicates terminal equipment Time slot Tasks processed in The amount of tasks;

[0016] Terminal task transfer queue:

[0017] ;

[0018] in, Indicates time slot Terminal equipment Tasks in the transmission queue Offloaded to edge servers The number of tasks;

[0019] Edge task computing queue:

[0020] ;

[0021] in, Indicates time slot Edge Server Task queue The number of tasks processed in It is a time slot Edge Server Upper container caching decisions;

[0022] S12. According to the characteristic parameters of the cloud server, edge server, terminal device and container involved in the edge computing environment, the processing time of the task on the terminal device or edge server, the transmission time of the task offloading from the terminal device to the edge server, the time when the edge server downloads the container image file from the cloud server, and the time when the container instance is started on the edge server are established into a mathematical model, the content of which is as follows:

[0023] Task execution time of terminal equipment:

[0024] ;

[0025] in, represents the computational complexity of the terminal device task, Indicates terminal equipment local computing power;

[0026] Task offloading time of terminal equipment:

[0027] ;

[0028] in, Indicates that from the terminal device To the edge server The task offloading rate;

[0029] Task execution time of edge server:

[0030] ;

[0031] in, Represents edge server computing power;

[0032] Container preparation time for edge servers:

[0033] ;

[0034] in, Represents a container The start time, represents the decision of caching image s on edge server j, belonging to {0, 1}, Represents a mirror The file size of Represents edge server The transfer rate of image files downloaded from the cloud server;

[0035] At the end of the time slot, the total remaining amount of all task queues in the system is:

[0036] ;

[0037] in, represents the offloading decision of terminal device i on task s, which belongs to {0, 1};

[0038] Total delay of all task processing:

[0039] .

[0040] Preferably, S2 specifically includes:

[0041] S21, optimize the objective function;

[0042] S22, optimization problem constraints, including terminal computing queue constraints, terminal transmission queue constraints, edge computing queue constraints, and time slot length constraints;

[0043] S23. Taking the weighted minimization of the task queue remaining amount and task processing delay in containerized edge computing as the optimization goal, a multi-slot nonlinear mixed integer programming problem is established by integrating the terminal computing queue, terminal transmission queue, edge computing queue, time slot length constraint, and queue stability constraint conditions.

[0044] Preferably, the optimization objective function in S21 is:

[0045] ;

[0046] in, Express expectations, Represents the weight sum function , Represents the weight value in the weight sum function.

[0047] Preferably, the optimization problem constraints in S22 are as follows:

[0048] Terminal calculation queue constraints:

[0049] .

[0050] Terminal transmission queue constraints:

[0051] .

[0052] Edge computing queue constraints:

[0053] .

[0054] Slot length constraints:

[0055] ;

[0056] ;

[0057] .

[0058] Preferably, in S3, the multi-slot problem is decomposed into a single-slot problem according to Lyapunov optimization theory:

[0059] ;

[0060] in, express , express , express , Represents the penalty term in the Lyapunov drift plus penalty function.

[0061] Preferably, the content of S4 is as follows:

[0062] S41. The original nonlinear mixed integer programming problem is equivalently converted into a linear mixed integer programming problem, as follows:

[0063] Introducing variables , and add constraints:

[0064] ;

[0065] in, express The maximum value obtained is the variable The upper bound of

[0066] Introducing variables , and add constraints:

[0067] ;

[0068] S42, calculating the parameters required for the linear mixed integer programming problem through an iterative calculation formula;

[0069] S43. Solve the linear mixed integer programming problem obtained after the transformation by using an existing solver or an existing linear integer programming algorithm.

[0070] Therefore, the present invention adopts the above-mentioned method to establish a joint optimization method of task offloading and resource allocation in containerized edge, establishes a mathematical model of containerized edge computing environment, and establishes an optimization problem that minimizes the task processing time of terminal devices. At the same time, a joint optimization algorithm for task offloading and resource allocation is proposed. This method transforms the originally difficult nonlinear mixed integer programming problem into an easy-to-solve linear mixed integer programming problem through an equivalent conversion method, solving the problem of mutual coupling between task offloading decisions and resource allocation decisions in edge computing. The algorithm can be used to quickly solve the joint optimization problem and reduce the task processing time of terminal devices in the containerized edge computing system.

[0071] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Figure 1 A container edge computing scenario diagram of a joint optimization method for task offloading and resource allocation in container edge computing provided by an embodiment of the present invention;

[0073] Figure 2 A queue modeling architecture diagram of a joint optimization method for task offloading and resource allocation in containerized edge computing provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0074] Example

[0075] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0076] Reference Figure 1 and Figure 2 The joint optimization method for task offloading and resource allocation in containerized edge computing proposed in this embodiment includes steps S1 to S4, wherein:

[0077] In step S1, a containerized edge computing system model is established;

[0078] Consider a three-layer containerized edge computing architecture: cloud server layer, edge server layer, and terminal device layer. Represents a collection of terminal devices, .gather Represents a collection of edge servers, .gather Represents a collection of images, .gather Represents the time slot set of the CEC system, .

[0079] By adopting the queue modeling method, terminal task computing queue and terminal task transmission queue are introduced in terminal devices, and edge task computing queue is introduced in edge servers.

[0080] (1) Terminal task calculation queue

[0081] ;

[0082] in, Indicates terminal equipment In time slot The tasks generated The amount of tasks, Indicates terminal equipment Mission In time slot The task offloading decision Indicates terminal equipment Time slot Tasks processed in the amount of tasks.

[0083] (2) Terminal task transmission queue

[0084] ;

[0085] in, Indicates time slot Terminal equipment in the transmission queue Task Offloaded to edge servers The number of tasks.

[0086] (3) Edge Task Computing Queue

[0087] ;

[0088] in, Indicates time slot Edge Server Task queue The number of tasks processed in It is a time slot Edge Server Upper container caching decisions.

[0089] According to the characteristic parameters of the cloud server, edge server, terminal device and container involved in the edge computing environment, a mathematical model is established for the processing time of the task on the terminal device or edge server, the time of unloading the task from the terminal device to the edge server, the time of the edge server downloading the container image file from the cloud server, and the time of starting the container instance on the edge server.

[0090] (1) Task execution time of terminal equipment;

[0091] ;

[0092] in, represents the computational complexity of the terminal device task, Indicates terminal equipment local computing power.

[0093] (2) Task offloading time of terminal devices;

[0094] ;

[0095] in, Indicates that from the terminal device To the edge server The task offloading rate.

[0096] (3) Task execution time of edge servers;

[0097] ;

[0098] in, Represents edge server computing power.

[0099] (4) container preparation time of edge servers;

[0100] ;

[0101] in, Represents a container The start time, represents the decision of caching image s on edge server j, belonging to {0, 1}, Represents a mirror The file size of Represents edge server The transfer rate of image files downloaded from the cloud server.

[0102] (5) The total remaining amount of all task queues in the system at the end of the time slot;

[0103] ;

[0104] in, represents the offloading decision of terminal device i on task s, which belongs to {0, 1};

[0105] (6) Total delay of all task processing;

[0106] .

[0107] In step S2, a multi-slot nonlinear mixed integer programming problem is established according to the mathematical model, as follows:

[0108] (1) Optimization objective function:

[0109] ;

[0110] in, Express expectations, Represents the weight sum function , Represents the weight value in the weight sum function.

[0111] (2) Optimization problem constraints:

[0112] a) Terminal calculation queue constraints:

[0113] ;

[0114] b) Terminal transmission queue constraints:

[0115] ;

[0116] c) Edge computing queue constraints:

[0117] ;

[0118] d) Slot length constraint:

[0119] ;

[0120] ;

[0121] .

[0122] (3) The entire optimization problem is established:

[0123] Taking the weighted sum minimization of the task queue remaining amount and task processing delay in containerized edge computing as the optimization goal, a multi-objective joint optimization problem is established considering constraints such as terminal computing queue, terminal transmission queue, edge computing queue, time slot length constraint, and queue stability.

[0124] In step S3, the multi-slot problem is decomposed into a single-slot problem, as follows:

[0125] According to Lyapunov optimization theory, the objective function is transformed into the following form:

[0126] ;

[0127] in, express , express , express , Represents the penalty term in the Lyapunov drift plus penalty function.

[0128] In step S4, the nonlinear terms in the original problem are linearized by equivalent transformation. The specific method is as follows:

[0129] First, the original nonlinear mixed integer programming problem can be equivalently converted into a linear mixed integer programming problem through linearization method.

[0130] Introducing variables , and add constraints:

[0131] ;

[0132] in, express The maximum value obtained, that is, the variable The upper bound of

[0133] Introducing variables , and add constraints:

[0134] .

[0135] Then, the parameters required for the linear mixed integer programming problem are calculated through an iterative calculation formula. Finally, the converted linear mixed integer programming problem is solved through existing solvers such as CPLEX, Gurobi, etc., or existing linear mixed integer programming algorithms such as branch and bound algorithms.

[0136] Therefore, the present invention adopts the above-mentioned joint optimization method of task offloading and resource allocation in containerized edge computing, establishes a mathematical model of a containerized edge computing environment, and establishes an optimization problem that minimizes the task processing time of terminal devices. At the same time, a joint optimization algorithm for task offloading and resource allocation is proposed. This method converts the originally difficult nonlinear mixed integer programming problem into an easy-to-solve linear mixed integer programming problem through an equivalent conversion method, solving the problem of mutual coupling between task offloading decisions and resource allocation decisions in edge computing. The algorithm can be used to quickly solve the joint optimization problem and reduce the task processing time of terminal devices in the containerized edge computing system.

[0137] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A joint optimization method for task offloading and resource allocation in containerized edge, characterized in that: include: S1. Establish a containerized edge computing system model, including a cloud server layer, an edge server layer, and a terminal device layer, and establish a mathematical model based on characteristic parameters of the cloud server, edge server, terminal device, and container involved in the edge computing environment; The specific contents are: S11. Using queue modeling, introduce terminal task computing queue and terminal task transmission queue in terminal devices, and introduce edge task computing queue in edge servers. The content is as follows: gather Represents a collection of terminal devices, ;gather Represents a collection of edge servers, ;gather Represents a collection of images, ;gather Represents the time slot set of the CEC system, ; Terminal task calculation queue: ； in, Indicates terminal equipment In time slot The tasks generated The amount of tasks, Indicates terminal equipment Mission In time slot The task offloading decision Indicates terminal equipment Time slot Tasks processed in The amount of tasks; Terminal task transfer queue: ； in, Indicates time slot Terminal equipment Tasks in the transfer queue Offloaded to edge servers The number of tasks; Edge task computing queue: ； in, Indicates time slot Edge Server Task queue The number of tasks processed in It is a time slot Edge Server Upper container caching decisions; S12. According to the characteristic parameters of the cloud server, edge server, terminal device and container involved in the edge computing environment, the processing time of the task on the terminal device or edge server, the transmission time of the task offloading from the terminal device to the edge server, the time when the edge server downloads the container image file from the cloud server, and the time when the container instance is started on the edge server are established into a mathematical model, the content of which is as follows: Task execution time of terminal equipment: ； in, represents the computational complexity of the terminal device task, Indicates terminal equipment local computing power; Task offloading time of terminal equipment: ； in, Indicates that from the terminal device To the edge server The task offloading rate; Task execution time of edge server: ； in, Represents edge server computing power; Container preparation time for edge servers: ； in, Representation container The start time, represents the decision of caching image s on edge server j, belonging to {0, 1}, Represents a mirror The file size of Represents edge server The transfer rate of image files downloaded from the cloud server; At the end of the time slot, the total remaining amount of all task queues in the system is: ； in, represents the offloading decision of terminal device i on task s, which belongs to {0, 1}; Total delay of all task processing: ； S2, establish a multi-slot nonlinear mixed integer programming problem based on the mathematical model; Specifically include: S21, optimization objective function; the optimization objective function is: ； in, Express expectations, Represents the weight sum function , Represents the weight value in the weight sum function; S22, optimization problem constraints, including terminal computing queue constraints, terminal transmission queue constraints, edge computing queue constraints, and time slot length constraints; S23. Taking the weighted sum minimization of the task queue remaining amount and task processing delay in containerized edge computing as the optimization goal, a multi-slot nonlinear mixed integer programming problem is established by integrating the terminal computing queue, terminal transmission queue, edge computing queue, time slot length constraint, and queue stability constraint conditions; S3, decompose the multi-slot problem into a single-slot problem; According to Lyapunov optimization theory, the multi-slot problem is decomposed into a single-slot problem: ； in, express , express , express , represents the penalty term in the Lyapunov drift plus penalty function; S4. Convert the nonlinear mixed integer programming problem into a linear mixed integer programming problem by equivalent transformation. The content is as follows: S41. The original nonlinear mixed integer programming problem is equivalently converted into a linear mixed integer programming problem, as follows: Introducing variables , and add constraints: ； in, express The maximum value obtained is the variable The upper bound of Introducing variables , and add constraints: ； S42, calculating the parameters required for the linear mixed integer programming problem through an iterative calculation formula; S43. Solve the linear mixed integer programming problem obtained after the transformation by using an existing solver or an existing linear mixed integer programming algorithm.

2. According to claim 1, a method for joint optimization of task offloading and resource allocation in containerized edge is characterized in that: The optimization problem constraints in S22 are as follows: Terminal calculation queue constraints: ； Terminal transmission queue constraints: ； Edge computing queue constraints: ； Slot length constraints: ； ； 。

Images