CN108509276A - Video task dynamic migration method in edge computing environment - Google Patents

Abstract

The invention discloses a method for dynamically migrating video tasks in an edge computing environment. In an ECN cluster in an edge environment, several ECNs are selected to run computing task scheduling migration schedulers, which are responsible for task distribution and scheduling, and other ECNs in the cluster undertake Tasks migrate jobs and provide computing resources. The computing task migration scheduler executes the task migration decision in each scheduling cycle, and decides whether the task is executed locally in the ECN or migrated to the data center. The tasks that need to be migrated to the data center are directly sent to the remote data center through the ECN transmission unit. The task execution unit performs analysis and processing. On the basis of satisfying network bandwidth and computing cost, the present invention dynamically schedules and processes tasks, realizes the goal of reducing the transmission delay of edge data and improving the task analysis and processing speed, thereby effectively ensuring low delay and high task migration process. Response, to improve the user's video service experience while achieving high video service quality.

Description

A Dynamic Migration Method of Video Tasks in Edge Computing Environment

technical field

The invention relates to a video task dynamic migration method in an edge computing environment, and belongs to the field of combining edge computing technology and video technology application.

Background technique

With the rapid development of network technology, we have entered the era of the Internet of Everything, and the number of network edge devices is increasing rapidly. According to the estimate of Cisco Global CloudIndex, by 2019, up to 45% of network data will be analyzed and processed at the edge of the network. And the resulting edge data is as high as Zettabyte level. The calculation and storage of the traditional cloud computing model are performed in a centralized manner in the data center. Under the background of the Internet of Everything, the centralized cloud computing model has the following problems to be solved when processing video services: 1) The linear growth of computing equipment in the data center cannot Efficiently process massive video data generated by network edge devices in a short period of time; 2) The transmission of unprocessed massive edge data to the data center increases the load of transmission bandwidth; 3) The centralized computing mode of the data center consumes a lot of energy. Therefore, an edge big data processing mode, namely edge computing, was born.

Edge computing refers to a new type of computing model that performs computing at the edge of the network, where the edge refers to any computing and network resources between data sources and cloud computing centers. As a service method of sharing computing resources, storage resources and application resources, edge computing nodes can provide customized computing services for other devices in a virtual computing environment. Edge computing migrates some or all of the original cloud-based computing tasks to network edge devices for execution, closer to the service objects of the tasks. In this way, the processing burden of the cloud computing center can be effectively reduced, and the pressure on network bandwidth can be relieved.

Migration of tasks is one of the technical difficulties of edge computing, which can be migrated between edge computing nodes and the cloud according to actual needs. Due to the unstructured nature of the video data itself, considering the influence of factors such as delay, computing power, and network bandwidth on task migration, the task execution strategies are different. Since the edge computing node is located near the data source, it can achieve lower latency and higher bandwidth, and migrate tasks that require high latency to the edge computing node, which can effectively improve the quality of video services and user experience; while computing Tasks that consume too much resources are migrated to the cloud for processing. The above task migration method is effective in reducing task transmission delay and balancing system energy consumption, but it does not consider task emergencies in specific business scenarios (such as video). At the same time, the concurrent task processing performance of the system also needs to be improved. Therefore, in the case of meeting the task delay requirements, how to design the migration strategy and means between the edge node and the cloud to reduce the consumption of network bandwidth for video task transmission and achieve user-satisfied service quality has become an academic issue. problems to be solved urgently by the world and the industry.

In order to reduce the impact of task transmission delay and energy consumption, Liyao Xiang et al. proposed a cluster migration strategy and designed an algorithm to migrate multiple tasks to the data center in a relatively energy-saving manner. energy consumption, but cannot meet the needs of low-latency and high-response video processing tasks during migration; You Lu et al. proposed a guiding forwarding scheme for content retrieval in edge computing environments, sharing hot video content to adjacent locations Other edge nodes can effectively reduce the communication overhead caused by duplicating the query copy. Due to the long distance between the edge node and the data center, the communication overhead and data processing overhead of transmitting hot video content for the first time are still high; Byung-Gon Chun et al. proposed A method of task cloning and migration. Through static analysis and dynamic analysis, tasks are split into multiple parts, and tasks are uniformly scheduled according to factors such as delay, energy consumption, and resource requirements. Although this method reduces communication overhead through task migration , however incurs additional local computational overhead when analyzing video tasks.

By comparing the progress and results of the above-mentioned task migration problem, it can be found that this kind of method is not ideal for the delay optimization effect of video services in the edge computing environment. The data center is close to the data source. In the edge computing environment, the impact of a long distance on the migration delay of video tasks cannot be ignored.

Contents of the invention

The technical problem to be solved by the present invention is to provide a delay-optimized video task dynamic migration method for the deficiencies of existing edge computing solutions in video services. Perform dynamic scheduling and processing to achieve the goal of reducing the transmission delay of edge data and increasing the speed of task analysis and processing, thereby effectively ensuring low delay and high response during task migration, and improving user experience while achieving high video service quality. Video service experience.

In order to solve the above-mentioned technical problems, the present invention specifically adopts the following technical solutions.

A method for dynamically migrating video tasks in an edge computing environment, comprising the steps of:

Step 001, monitor the resource consumption status of each edge computing node ECN and the status information of the video task in the edge computing cluster;

Step 002, ECN generates a task queue according to the current video task status information, and the computing task migration scheduler in the edge computing cluster compares the computing resource consumption status of each task, the current network bandwidth situation with the preset threshold, and determines whether to assign the task to transmission unit.

Step 003: The computing task migration scheduler randomly arranges the tasks in the transmission unit, and makes a migration decision based on the principle of minimizing time delay; the transmission unit migrates the video tasks within the scheduling period to the local ECN or migrates them to a remote location according to the migration decision Cloud Computing Center Execution.

Further, in the method for dynamically migrating video tasks in an edge computing environment proposed by the present invention, the resource consumption status of ECN specifically includes: CPU utilization, memory usage, and cache space; the status information of the video processing task includes type, Data volume, task completion deadline.

Furthermore, in the method for dynamically migrating video tasks in an edge computing environment proposed by the present invention, the computing resource consumption status described in step 002 is generated in real time according to the conditions of each video task, specifically: set the total computing resources of the ECN cluster is μ _e , the maximum computing resource available in the data center is μ _c , then the maximum computing resource allowed to process the migration task is μ _max ≥ μ _e + μ _c ; let the computing resource consumed by task i during the migration process be μ _i , Then for a queue with N tasks, the following conditions must be met for successful migration in each scheduling cycle: That is, the sum of computing resources consumed by task migration is less than or equal to the maximum computing resources provided by the ECN cluster and data center.

Further, in the method for dynamically migrating video tasks in an edge computing environment proposed by the present invention, the preset threshold of network bandwidth in step 002 is generated according to the channel bandwidth in the edge environment where each ECN is located, specifically: when satisfying In the case of β×log(1+φ _i |h _i | ² )≥r _i , the channel bandwidth of ECN is dynamically adjusted for the migration of computing tasks, where β is the bandwidth allocated by ECN for tasks, and φ _i is the The SNR of the i-th task, |h _i | ² is the channel gain, and ri is the minimum transmission rate of the task.

Further, in the method for dynamic migration of video tasks in an edge computing environment proposed by the present invention, in step 003, the task migration decision indicator is expressed as δ _e [t], δ _c [t] ∈ {0, 1}, 0 means no migration, otherwise 1 means execution task migration; δ _e [t] and δ _c [t] represent the migration decision of the edge node ECN and the computing center respectively;

At time t, the following task queues exist:

ρ[t+1]=ρ[t]-δ _e [t]-δ _c [t]+α _e [t], t=1,2,3... (1)

Among them, α _e [t] ∈ {0, 1} indicates whether there are new computing tasks arriving in the queue at time t, if a new task arrives α _e [t] = 1, otherwise α _e [t] = 0; transfer the task The state is expressed as Θ={(δ _e [t],δ _c [t])|(0,1),(1,0),(1,1),(0,0)}, there are the following four transitions decision making:

(1), Under this migration decision, both the ECN and the data center task queue are idle, and at least two tasks can be executed at the same time;

(2), Under this migration decision, the ECN task queue is idle, and the data center task queue is scheduling tasks, and tasks can be migrated to ECN;

(3), Under this migration decision, the ECN task queue is scheduling tasks, and the data center task queue is idle, and tasks can be migrated to the data center;

(4), Under this decision, both the ECN and the data center task queue are executing scheduling tasks, and tasks cannot be migrated temporarily.

Further, a dynamic video task migration method in an edge computing environment proposed by the present invention uses Indicates the number of time slices required by the ECN to execute local task i at time t, where n is the number of time slices that have already been run, and N represents the total number of time slices required to execute video processing task i, L _i is the data volume of task i, 1≤n≤N-1;

According to the defined task migration status, the Indicates the number of time slices required by the data center to execute task i at time t. A task with a data volume of C needs to be divided into multiple data packets for execution, where 1≤m≤R, and R is the data packet required for task C to be divided. number.

Further, in the method for dynamically migrating video tasks in an edge computing environment proposed by the present invention, in step 003, the migration of computing tasks is aimed at minimum delay optimization, so as to ensure the real-time performance and high response of video services; Each computing task undergoes two states of waiting and executing during the migration process; according to Little’s law, there is L=λ×W, where λ is the task arrival rate, and W represents the average execution time of the task; during the task migration process, both The execution time of ECN also has the execution time of the data center, so the task processing delay is:

t _p = λ×t _e +(1-λ)×t _c , λ∈[0,1] (2)

Among them, λ represents the proportion of computing tasks executed in ECN, then the total task migration delay can be expressed as:

T＝t _p +t _q +t _t +t _f (3)

If the latest completion time of the i-th task in the task queue is τ _i , the criterion for successful task migration is that the sum of transmission and execution delays is less than τ _i , that is, the following relationship exists:

p _i ＝P(t _q +t _p +t _t +t _f ≤τ _i ) (4)

Among them, t _e represents the execution time of ECN, t _c represents the execution time of the data center, t _t is the transmission delay of task i, t _f is the feedback delay of task i after processing is completed; t _q represents the average queue waiting delay .

Furthermore, in the method for dynamic migration of video tasks in an edge computing environment proposed by the present invention, the total time delay of task migration during the dynamic migration process can be transformed into the following problem:

Among them, μ _e is the total computing resources of the ECN cluster, μ _c is the computing resources available in the data center, μ _max is the maximum computing resources allowed to process migration tasks, β is the bandwidth allocated by ECN for tasks, and φ _i is the number of tasks in the task queue The SNR of the i-th task, |h _i | ² is the channel gain, ri is the minimum transmission rate of the task, and ρ _ζζ' is expressed as the one-step transition probability from state ζ to ζ', there is the following relationship

Compared with the prior art, the present invention adopts the above technical scheme and has the following technical effects:

1. Apply the Markov chain method to construct a random computing task scheduling model, obtain the calculation method of the delay in each stage of the task execution and transmission process, and theoretically obtain the accurate value of the analysis processing overhead and transmission overhead when the video task is migrated.

2. Divide the migration strategy of computing tasks, and the method can be applied to large-scale and highly concurrent distributed task scheduling systems.

3. Analyze the processing status of video tasks. In the case of large traffic, it can improve the utilization rate of network bandwidth, reduce the overhead of computing resources in the data center, smooth the emergence of sudden tasks, and improve the concurrency of the system.

4. The optimization goal is to minimize the total time delay during video task migration. The migration effect for video tasks is better, with the characteristics of low time delay and high response.

Description of drawings

Fig. 1 is a frame diagram involved in the dynamic migration method of video tasks in the edge computing environment designed by the present invention.

Fig. 2 is a diagram of migration steps in the method for dynamic migration of video tasks in the edge computing environment designed by the present invention.

Detailed ways

The specific implementation manners of the present invention will be further described in detail below in conjunction with the accompanying drawings.

Those skilled in the art can understand that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It should also be understood that terms such as those defined in commonly used dictionaries should be understood to have a meaning consistent with the meaning in the context of the prior art, and unless defined as herein, are not to be interpreted in an idealized or overly formal sense explain.

In the edge computing environment, since the geographical location of the ECN cluster is closer to the data source, it can achieve lower latency, higher response, and higher bandwidth. Video tasks with high latency requirements are migrated to edge computing nodes to complete, which can effectively improve services. quality and user experience.

The video task dynamic migration method in the edge computing environment proposed by the present invention performs task migration and scheduling with the goal of minimizing time delay within the allowable range of computing power and network bandwidth, and makes full use of the computing resources and storage capabilities of the ECN cluster. In the process of task migration, there are mainly two problems: whether to perform task migration and how to allocate resources to process tasks. Considering that task migration is affected by various factors such as delay, network bandwidth, energy consumption, and computing power, the dynamic task migration method is mainly suitable for computing tasks with high delay requirements (such as streaming media services, mobile image processing, etc.) , it is better for low-latency, high-response task migration.

As shown in Figure 1, the present invention designs a dynamic migration method for video tasks in an edge computing environment. The edge computing environment is mainly composed of an ECN cluster close to the data source, and the ECN cluster and the remote data center jointly provide computing services for users. , make reasonable use of computing resources through task migration. The ECN used in the present invention is a small-scale computing platform. The virtual machine allocated by the ECN runs the program of task migration scheduling, and also provides CPU computing resources during the task migration process, and has a certain storage capacity at the same time.

A method for dynamically migrating video tasks in an edge computing environment, comprising the steps of:

Step 001. Obtain the status of each edge node (Edge Compute Node, ECN) and the video processing task to be executed in the edge computing cluster, specifically including the resource consumption status of each ECN (CPU utilization rate, memory usage rate, cache space, etc.), video Process task status information (type, data volume, task completion deadline, etc.).

Step 002. The ECN generates a task queue (Task Queue, TQ) according to the status information of the current video task, and the number of cacheable tasks in TQ is Q. The tasks in TQ can be migrated to ECN for execution, or transferred to the data center for execution. The task migration decision indicator is expressed as δ _e [t], δ _c [t] ∈ {0, 1}, 0 means no migration, otherwise 1 means execution of task migration, and the task migration state is expressed as Θ={(δ _e [t],δ _c [t])|(0,1),(1,0),(1,1),(0,0)}. At time t, the following task queues exist:

ρ[t+1]=ρ[t]-δ _e [t]-δ _c [t]+α _e [t], t=1,2,3... (1)

Among them, α _e [t]∈{0, 1} indicates whether a new computing task arrives in the queue at time t, if a new task arrives α _e [t]=1, otherwise α _e [t]=0.

Step 003. The computing task migration scheduler randomly arranges the tasks in the transmission unit, and makes a migration decision based on the principle of minimizing time delay. ECN executes task scheduling decisions in each time period t, so as to decide whether tasks are migrated to ECN for execution or transmitted to the data center for execution. According to (δ _e [t], δ _c [t]) defined in step 002, there are the following four migration decisions:

1, Under this migration decision, both the ECN and the data center task queue are idle, and at least two tasks can be executed at the same time;

2, Under this migration decision, the ECN task queue is idle, and the data center task queue is scheduling tasks, and tasks can be migrated to ECN;

3. Under this migration decision, the ECN task queue is scheduling tasks, and the data center task queue is idle, and tasks can be migrated to the data center;

4. Under this decision, both the ECN and the data center task queue are executing scheduling tasks, and tasks cannot be migrated temporarily.

According to the migration decision, the transmission unit migrates the video tasks within the scheduling period to be executed locally in the ECN or to a remote cloud computing center for execution. Each computing task goes through two states (waiting state, execution state) during the migration process. According to Little's law, there is L=λ×W. Among them, λ is the arrival rate of the task, and W is the average execution time of the task. During the task migration process, there are both ECN execution time and data center execution time, so the task processing delay is:

t _p = λ×t _e +(1-λ)×t _c , λ∈[0,1] (2)

Among them, λ represents the proportion of computing tasks performed in ECN. Then the total task migration delay can be expressed as:

T＝t _p +t _q +t _t +t _f (3)

t _t is the transmission delay of task i, and t _f is the feedback delay after task i is processed.

As a preferred technical solution of the present invention: the step 001 also includes task arrival analysis, considering that the number of task arrivals in [0, T] of the task queue in the edge environment is N(t), which is a stable independent increasing Quantitative process, the task arrival obeys the Poisson flow {N(t), t≥0} with parameter λ.

As a preferred technical solution of the present invention: the task execution time in step 002 is divided into equal time slices τ, and the execution time of the task migration strategy is an integer multiple of τ. At the same time, the network bandwidth for task migration is limited, and the bandwidth allocated by ECN to the task is β. For the i-th task in the task queue, the SNR is φ _i , and the channel gain is |h _i | ² . It is equal to the minimum transmission rate ri of the task, that is, the following conditions are met: β×log(1+φ _i |h _i | ² )≥r _i , and the network bandwidth preset threshold is generated according to the channel bandwidth in the edge environment where each ECN is located. Dynamically adjust the channel bandwidth of ECN for the migration of computing tasks.

As a preferred technical solution of the present invention: the step 002 also includes consideration of computing resources during task migration, and feedback of real-time computing resource consumption status according to each ECN and data center hardware usage. Suppose the total computing resource of the ECN cluster is μ _e , and the maximum computing resource available in the data center is μ _c , then the maximum computing resource allowed for task execution is μ _max ≥ _{μ e} + μ _c , and the computing resource consumed during the migration of task i The resource is μ _i , then the following conditions must be satisfied for the successful migration of a queue with N tasks: That is, the sum of computing resources consumed by task migration is less than or equal to the maximum computing resources provided by the ECN cluster and data center.

As a preferred technical solution of the present invention: the step 003 also includes dynamically assigning tasks to migrate to idle ECNs according to the load and utilization of each node in the ECN cluster, balancing the computing power of each node in the ECN cluster and utilizing them reasonably Network bandwidth to achieve the goal of efficient use of network and hardware resources. The ECN task execution unit is responsible for computing tasks that run locally on the ECN. In order to conveniently measure the task migration status of each ECN, use Indicates the number of time slices that ECN still needs to execute local task i at time t (n is the number of time slices that have already been run), where, Indicates the total number of time slices required to execute the video processing task i whose data size is L _i . The cloud computing task migration unit is responsible for performing tasks that are migrated to the cloud server for execution. According to the task migration status defined in step 003, the Indicates the number of time slices required by the data center to execute task i at time t, and a task with a data volume of C needs to be divided into multiple data packets for execution, where R is the number of data packets required for task C to be divided.

As a preferred technical solution of the present invention: the step 003 also includes that the latest completion time of the i-th task in the task queue is τ _i , and the judging condition for successful task migration is that the sum of transmission and execution delays is less than τ _i , that is, the following relationship exists:

p _i ＝P(t _q +t _p +t _t +t _f ≤τ _i ) (2)

where p _i represents the probability of successful transfer of task i.

As shown in Figure 2, it is a detailed step diagram of the video task dynamic migration method, explaining the main steps and data interaction process in the video task dynamic migration method. The English terminology is explained as follows:

Client: data source or client, providing calculation tasks and rendering task migration results;

Router: In the edge environment, reasonably allocate the network bandwidth of each ECN according to the network environment;

Data Center: The task of computing migration to the data center, but the distance from the data source is far away, and the transmission delay is high;

Edge Computing Node: the edge computing node ECN, the ECN used in the present invention is a small computing platform, running the program of task migration and scheduling in the virtual machine of ECN, providing CPU for task processing, and having a certain storage capacity at the same time.

In the ECN cluster in the edge environment, several ECNs are selected to run computing task scheduling and migration schedulers, which are responsible for task distribution and scheduling. Other ECNs in the cluster are mainly responsible for task migration and providing computing resources. The computing task scheduler first obtains the resource consumption status and task status information of each ECN in the cluster, and balances the impact of network bandwidth and ECN hardware on service quality and user experience during task migration. Compare the ECN and task status obtained by monitoring with the preset threshold value (the threshold value is artificially preset during the initial operation, and the threshold value can be updated according to the adjustment of the network environment and hardware resources or the optimal threshold value of each ECN can be determined by calculation), Determine the number of ECNs that can perform task migration, traverse the tasks to be migrated in the cluster, and allocate computing tasks according to the results of each ECN's computing resource consumption status and network bandwidth status feedback.

The computing task migration scheduler executes the task migration decision in each scheduling period t, and decides whether the task i is to be executed locally in the ECN or migrated to the data center. The tasks that need to be migrated to the data center are directly sent to the remote data via the ECN transmission unit The central task execution unit performs the analysis and processing. In order to minimize the time delay of task migration and facilitate the execution of migration decisions, a Markov chain is introduced to describe the random computing task scheduling model, using triplets Indicates the state of ζ _p [t], and the corresponding state space S can be expressed as: S={0,1,2,...,Q}×{0,1,2,...,R}×{0 ,1,2,...,N-1}. Express ρ _ζζ' as the one-step transition probability from state ζ to ζ', let the stationary distribution defined in S be There is the following relationship

By introducing the probability parameter The decision state can be mapped into a probability space, where Represents the probability of executing each migration decision; δ=1,2,3,4 represents four kinds of migration decisions for task i, namely {(δ _e [t],δ _c [t])|(0,1),( 1,0),(1,1),(0,0)}, the task i can be executed locally in the ECN, or can be migrated to the data center for execution. When the delay and computing resources allow, the video task i can be The split is performed in both places simultaneously.

The computing task migration scheduler allocates the tasks to be migrated within the scheduling period t to the task execution units of each ECN and data center according to the principle of minimizing the delay. This principle gives priority to the higher channel bandwidth or more computing resources in the ECN cluster In order to ensure that the task can be completed before the delay deadline after migration, and at the same time balance the task load of each ECN in the edge environment, there are the following four migration decisions:

1, In this scheduling scenario, both the ECN and the data center task queue are in an idle state. The computing task migration scheduler can dispatch video migration tasks to the two places, and can process at least two computing tasks at the same time. The state under this decision can be expressed as Including the following four migration strategies {(δ _e [t],δ _c [t])|(0,1),(1,0),(1,1),(0,0)}, the corresponding probabilities are respectively for Considering the arrival of tasks in the queue comprehensively, the initial state is The task arrival rate is α, and the corresponding transition probability is as follows:

2, In this scheduling scenario, the task execution unit of the ECN is idle, and the task execution unit of the data center is processing the task. The computing task migration scheduler assigns task i to the ECN for local processing. There are two migration strategies as follows {(δ _e [t], δ _c [t])|(0,1),(0,0)}, the corresponding probabilities are The initial state is The corresponding transition probabilities are as follows:

3. In this scheduling scenario, the task execution unit of the ECN is processing the task, and the task execution unit of the data center is in an idle state. The computing task migration scheduler can dispatch computing tasks to the data center. There are two migration strategies as follows {(δ _e [t], δ _c [t])|(1,0),(0,0)}, the corresponding probabilities are The initial state is The corresponding transition probabilities are as follows:

4. In this scheduling scenario, both the ECN and the data center task execution unit are processing tasks, and the computing task migration scheduler is temporarily unable to dispatch migration tasks in this scheduling period t, and the migration strategy is {(δ _e [t],δ _c [t]) |(0,0)} with probability

The task execution time is divided into equal time slices τ, and the execution time of the task migration strategy is an integer multiple of τ. Each computing task needs to go through two stages (waiting state, execution state) during its operation. In the waiting state, that is, the waiting delay of tasks in the queue can be expressed as:

in, During the task migration process, there are both the execution time of the ECN and the execution time of the data center, so the task processing delay t _p = max{t _e ,t _c }, and the total task migration delay is T=t _p +t _q +t _t +t _f , where t _q is the waiting delay of task i in the queue, t _t is the transmission delay of task i, and t _f is the feedback delay of task i after processing. Therefore, the total delay of task migration during dynamic migration can be transformed into the following problem:

The delay optimization problem is a linear programming, which is convenient to obtain the total delay of task migration under the current environment.

The video task dynamic migration method used in the edge computing environment also has two modes of service objects: single user and multi-user, that is, whether the provider of the video source is the only source or multiple sources, and the corresponding video task migration strategy also has different. In the multi-user service mode, in addition to considering the minimum delay, the resource consumption during multi-user task migration also needs to be considered. How to allocate computing resources reasonably is also an urgent problem to be solved. Therefore, in the single-user Based on the stochastic computing task scheduling model, such problems can be solved by evenly allocating network bandwidth and computing resources.

The present invention combines the two important factors of task migration delay and migration computing resources in the edge computing environment to propose a delay-optimized video task dynamic migration method. Features, the impact of transmission delay and processing delay on task completion time during video task migration cannot be ignored, and the present invention combines the impact of migration delay on video service quality and the additional calculation Balanced resource consumption is considered uniformly, and ECN tasks are dynamically and jointly scheduled for metrics such as power consumption, task migration delay, and computing resources of the ECN cluster, which can effectively improve the analysis and processing speed of video tasks and ensure low latency and high response of task migration .

Those skilled in the art will understand that the present invention may relate to an apparatus for performing one or more of the operations described in this application. Said apparatus may be specially designed and fabricated for the required purposes, or it may comprise known apparatus in a general purpose computer selectively activated or reconfigured by a program stored in it. Such a computer program can be stored in a device (e.g., computer) readable medium, including but not limited to any type of medium suitable for storing electronic instructions and respectively coupled to a bus. Types of disks (including floppy disks, hard disks, compact disks, CD-ROMs, and magneto-optical disks), random access memory (RAM), read-only memory (ROM), electrically programmable ROM, electrically erasable ROM (EPROM), electrically erasable Programmable ROM (EEPROM), flash memory, magnetic card or optical card. Readable media include any mechanism for storing or transmitting information in a form readable by a device (eg, a computer). Readable media include, for example, random access memory (RAM), read only memory (ROM), magnetic disk storage media, optical storage media, flash memory devices, signals propagated in electrical, optical, acoustic, or other forms (such as carrier waves, Infrared signal, digital signal), etc.

It will be understood by those skilled in the art that computer program instructions can be used to implement each block in these structural diagrams and/or block diagrams and/or flow diagrams and the blocks in these structural diagrams and/or block diagrams and/or flow diagrams The combination. These computer program instructions may be provided to a general-purpose computer, specialized computer, or other programmable data-processing processor to create a machine, whereby the instructions executed by the computer or other programmable data-processing processor create a structure for implementing A method specified in a box or boxes of a diagram and/or a block diagram and/or a flow diagram.

Those skilled in the art can understand that the various operations, methods, and steps, measures, and solutions in the processes discussed in the present invention can be replaced, changed, combined, or deleted. Further, other steps, measures, and schemes in the various operations, methods, and processes that have been discussed in the present invention may also be replaced, changed, rearranged, decomposed, combined, or deleted. Further, steps, measures, and schemes in the prior art that have operations, methods, and processes disclosed in the present invention can also be alternated, changed, rearranged, decomposed, combined, or deleted.

The embodiments of the present invention have been described in detail above in conjunction with the accompanying drawings, but the present invention is not limited to the above embodiments, within the knowledge of those of ordinary skill in the art, you can also Make various changes.

Claims (8)

1. A method for dynamically migrating a video task in an edge computing environment is characterized by comprising the following steps:

001, monitoring the resource consumption state of each edge computing node ECN and the state information of a video task in the edge computing cluster;

step 002, the ECN generates a task queue according to the current video task state information, and the computation task migration scheduler in the edge computation cluster compares the computation resource consumption state of each task and the current network bandwidth condition with a preset threshold value, and determines whether to allocate the task to the transmission unit.

Step 003, calculating a task migration scheduler to randomly arrange tasks in the transmission unit, and making a migration decision according to a principle of minimizing time delay; and the transmission unit migrates the video tasks in the scheduling period to be executed locally in the ECN or to be executed by a remote cloud computing center according to the migration decision.

2. The method according to claim 1, wherein the resource consumption status of the ECN specifically includes: CPU utilization rate, memory utilization rate and cache space; the state information of the video processing task comprises the type, the data volume and the task completion deadline.

3. The method according to claim 1, wherein the computing resource consumption status in step 002 is generated in real time according to the situation of each video task, and specifically comprises: let the total computing resources of the ECN cluster be μ_eData center can use up to mu of computing resources_cThen the maximum computing resource allowed for processing the migration task is μ_max≥μ_e+μ_c(ii) a Let the computational resource consumed by task i during migration be μ_iThen, the following condition is required to be satisfied for the queue with N tasks to migrate successfully in each scheduling period:that is, the sum of the computing resources consumed by the task migration is less than or equal to the maximum computing resources provided by the ECN cluster and the data center.

4. the method of claim 1, wherein the network bandwidth pre-set threshold of step 002 is generated according to the channel bandwidth difference in the edge environment of each ECN, and is defined as β × log (1+ φ) when the requirement is satisfied_i|h_i|²)≥r_iin the case of (1), dynamically adjusting the channel bandwidth of the ECN for the migration of the computing task, where β is the bandwidth allocated by the ECN for the task, φ_iIs the SNR, | h, of the ith task in the task queue_i|²For channel gain, ri is the mission minimum transmission rate.

5. The method for dynamically migrating video task in edge computing environment according to claim 1, wherein in step 003, the task migration decision indicator is represented as δ_e[t],δ_c[t]E is left to {0, 1}, 0 is not migrated, otherwise 1 is executed task migration; delta_e[t]And delta_c[t]Respectively representing migration decisions of the edge node ECN and the computing center;

at time t, there are the following task queues:

ρ[t+1]＝ρ[t]-δ_e[t]-δ_c[t]+α_e[t],t＝1,2,3... (1)

wherein alpha is_e[t]e {0, 1} represents whether a new calculation task arrives at the queue at the time t or not, and if the new task arrives at the alpha_e[t]1, otherwise_e[t]0; representing the task migration state as Θ { (δ)_e[t],δ_c[t]) L (0,1), (1,0), (1,1), (0,0) }, there are four migration decisions:

(1)、the ECN and the data center task queue are both in an idle state under the migration decision, and at least 2 tasks can be executed simultaneously;

(2)、the task queue of the ECN is in an idle state under the migration decision, and the data center task queue is scheduling tasks and can migrate the tasks to the ECN;

(3)、task of ECN under the migration decisionThe queue is scheduling the task, the data center task queue is in the idle state, can move the task to the data center;

(4)、under the decision, both the ECN and the data center task queue execute the scheduling task, and the task cannot be migrated temporarily.

6. The method of claim 1, wherein the dynamic migration of video tasks in an edge computing environment is performed byRepresents the number of time slices required by the ECN to execute the local task i at t time, wherein N is the number of time slices already running, N represents the total number of time slices required by executing the video processing task i,L_in is more than or equal to 1 and less than or equal to N-1;

according to the defined task migration condition, willThe method comprises the steps that the number of time slices required by a data center to execute a task i at the time t is represented, the task with the data volume of C needs to be divided into a plurality of data packets to be executed, wherein m is larger than or equal to 1 and is smaller than or equal to R, and R is the number of the data packets required to be divided by the task C.

7. The method for dynamically migrating video tasks in an edge computing environment according to claim 6, wherein in step 003, migration of the computing tasks is targeted at minimum delay optimization, so as to ensure real-time performance and high response of the video service; each computing task undergoes two states of waiting and executing during the migration process; according to the litter's law, L ═ λ × W, where λ is the task arrival rate and W represents the average task execution time; in the task migration process, there are the execution time of the ECN and the execution time of the data center, so the task processing delay is:

t_p＝λ×t_e+(1-λ)×t_c，λ∈[0,1](2)

where λ represents the proportion of the computational tasks performed in the ECN, the total task migration latency can be expressed as:

T＝t_p+t_q+t_t+t_f(3)

if the latest completion time of the ith task in the task queue is tau_iIf the task migration success is judged as the sum of the transmission delay and the execution delay is less than tau_iThat is, the following relationship exists:

p_i＝P(t_q+t_p+t_t+t_f≤τ_i) (4)

wherein, t_eRepresenting the execution time, t, of the ECN_cRepresenting the execution time, t, of the data center_tFor the transmission delay of task i, t_fThe feedback time delay after the task i is processed is obtained; t is t_qRepresenting the average queue latency.

8. The method for dynamically migrating the video task in the edge computing environment according to claim 7, wherein the total task migration delay in the dynamic migration process can be converted into the following problem:

wherein, mu_eTotal computational resources, μ, for ECN clusters_cMu computing resources available to the data center_maxβ is the bandwidth, phi, allocated by the ECN to the task in order to process the maximum computing resource allowed by the migration task_iIs the SNR, | h, of the ith task in the task queue_i|²To channel gain, r_iFor the task minimum transmission rate, let ρ_ζζ'The probability of one-step transition, expressed as state ζ to ζ', has the following relationship