Cloud edge-end function calculation unloading method capable of cooperatively transmitting data

Abstract

The invention discloses a cloud edge-end function calculation unloading method capable of cooperatively transmitting data. According to the method, a function calculation unloading process is divided into a data transmission process and a function calculation process. The function calculation process depends on data transmission among the nodes, so that a collaborative data collaboration mechanism can be introduced in the data transmission process, and the task calculation time delay is further reduced. Compared with the prior art, the method has the advantages that a data unloading mechanism is introduced in the cloud side-end collaborative computing unloading process under the complex function dependency relationship; task calculation time delay is reduced, and calculation efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the fields of cloud computing, computing offloading and data transmission, and in particular to a cloud-edge function computing offloading method capable of collaborative data transmission. Background Art

[0002] Computation offloading and data transmission are two very important collaborative technologies in cloud-edge-end collaboration. In terms of computation offloading, due to the development of artificial intelligence, computation-intensive applications such as pattern recognition, interactive games, virtual / augmented reality, etc., not only consume a lot of computing resources and energy, but also require ultra-low response latency. However, due to the limitations of hardware technology, the computing resources, storage resources, and energy resources of mobile devices are constrained to varying degrees, and cannot efficiently support the implementation of the above-mentioned delay-sensitive, computation-intensive applications. Therefore, it is considered to offload the user's computing tasks to the cloud data center for processing, and use the sufficient computing, storage and other resources of the cloud server to meet the needs of mobile applications for various resources. In the context of the Internet of Things and big data, the generation of massive data and the emergence of large-scale applications have put forward strict requirements on server memory and network uplink and downlink bandwidth, and limit the computing performance of terminal applications. The migration or offloading of huge task data and application data to the cloud server not only puts great pressure on bandwidth, but also brings severe tests to the user's service quality. Due to factors such as network instability and routing uncertainty, high transmission latency and poor reliability will occur, which seriously affects the efficiency of task computing. Therefore, cloud computing has shortcomings in supporting delay-sensitive applications and achieving millisecond-level latency.

[0003] At present, the development of computing offloading technology has gone through three stages, namely, complete task offloading, partial task offloading, and dependent task offloading. Complete offloading offloads all computing tasks to the cloud-edge-end collaborative network; partial offloading offloads part of the computing tasks to the cloud-edge-end collaborative network, and the terminal device and the cloud computing or edge computing nodes jointly calculate the tasks, and finally merge the data at the terminal. Complete offloading focuses more on the rapid execution of tasks on a high-performance node, while partial offloading focuses on how to allocate tasks. These two computing offloading strategies cannot actually truly utilize the resource advantages of the cloud-edge-end collaborative network. The computing tasks offloaded to the edge computing nodes or cloud computing nodes can only use limited resources, and cannot give full play to the role of cloud-edge-end collaboration.

[0004] In recent years, the problem of task-dependent computing offloading has been raised and gradually attracted attention, providing a new idea for exerting the cloud-edge-end collaboration capabilities. The basic idea is to construct task dependencies for many small tasks in a large task through DAG, and perform computing offloading based on the data forwarding relationship between tasks. In this way, a large computing task can be offloaded to more nodes for computing, which can better exert the advantages of cloud-edge-end collaboration compared to traditional computing offloading.

[0005] However, many studies have found that both traditional computing offloading methods and task-dependent methods require a large amount of program data to be uploaded during the computing offloading process, and a large number of identical computing tasks are repeatedly offloaded. Since data transmission and task code uploading are required during the offloading process, it will cause huge network transmission pressure, and the same computing tasks will repeatedly occupy computing and storage, resulting in a large amount of resource waste. Summary of the invention

[0006] Purpose of the invention: In view of the problems existing in the prior art, the present invention provides a cloud-edge function computing offloading method capable of collaborative data transmission, which divides the function computing offloading process into a data transmission process and a function computing process. The function computing process relies on data transmission between nodes, so it is possible to consider introducing a collaborative data offloading mechanism in the data transmission process, thereby further reducing the task computing latency. Therefore, a joint design is made to introduce a data offloading mechanism in the computing offloading process of cloud-edge collaboration under complex functional relationships. Compared with traditional methods and function computing offloading methods, this design can reduce the task computing latency by about 10%.

[0007] Technical solution: To achieve the purpose of the present invention, the technical solution of the present invention is: a cloud-edge function computing offloading method capable of data collaborative transmission, combining the data collaborative transmission mechanism with function computing offloading, including the following steps:

[0008] Step 1: In the cloud-edge-end collaboration framework, the end user uses the function computing mode to offload tasks to the cloud-edge collaborative network;

[0009] Step 2: Determine the optimization objective function, calculate the offloading location of each function, determine the nodes involved in the task calculation and transmit the original data to the first function;

[0010] Step 3: Find the multi-input and multi-output function dependencies in the offloaded functions, and use the data coordination mechanism to enhance data transmission capabilities, reduce data latency, and also reduce the computing latency of the task.

[0011] Compared with the prior art, the present invention has the following beneficial effects:

[0012] (1) The present invention proposes a function-level computing offloading method to solve the high latency problem of uploading traditional computing offloading program tasks.

[0013] (2) A collaborative data offloading mechanism is introduced into the data transmission process in the present invention, thereby reducing task computing delay and improving computing efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 Offload process for function calculation;

[0015] Figure 2 Grouping process for nodes;

[0016] Figure 3 To calculate the priority process;

[0017] Figure 4 For the data transmission process. DETAILED DESCRIPTION

[0018] The terms used in the present invention are only for the purpose of illustrating the embodiments of the present invention, and are not intended to limit the present invention. Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0019] The present invention proposes a function-level computing offloading method, which solves the high latency problem of uploading traditional computing offloading program tasks. At the same time, it introduces collaborative data transmission to solve the problem of high data transmission latency between nodes in the process of computing offloading with functions as the offloading target, under the condition of multi-input and multi-output data transmission relationship.

[0020] The technical solution adopted by the present invention is a joint method of data collaboration and function computing offloading, and the specific steps are as follows:

[0021] Step 1: Under the cloud-edge-end collaboration framework, end users use the function computing mode to offload tasks to the cloud-edge collaborative network to improve the computing power of the tasks;

[0022] First, it is necessary to obtain the status information of each node in the network, including node computing power, inter-node bandwidth, and number of nodes. The terminal needs to confirm its own task information and split the task into several functions, determine the input-output relationship between functions, and use a directed acyclic graph to represent the dependency relationship between functions within the task.

[0023] The specific steps of the function calculation mode and the method of determining the input-output relationship between functions and using a directed acyclic graph to represent the function dependency relationship within a task are as follows:

[0024] Step 1.1: Deploy a FaaS (cloud computing) platform on each node in the cloud-edge network to provide function computing service capabilities and build a function library that contains all the functions required for terminal tasks;

[0025] Step 1.2: Obtain network parameters and obtain the information of each node and task in the cloud-edge network through data interaction, where the node information includes the node computing capacity f i , the bandwidth B between nodes, the node number n in the link, and the task information includes the number of functions m that make up the task and the amount of computation w required for each function.

[0026] Step 1.3: Construct a DAG (directed acyclic graph) based on the decomposition relationship of the tasks, expressed as G(V,E), where V represents the function set of the graph vertices and E represents the edges of the graph, which is used to clarify the connection relationship between functions;

[0027] Step 2: Determine the optimization objective function, calculate the offloading location of each function, determine the nodes involved in the task calculation and transmit the original data to the first function. In the above process, the functional relationship of the task can be varied and can be artificially formulated, and different functions will be deployed on different cloud computing or edge computing nodes;

[0028] The specific steps of determining the optimization objective function and calculating the unloading position of each function are as follows:

[0029] Step 2.1: Establish an optimization objective function; the optimization objective of function computing is to minimize the computational latency of the task, that is, the end-to-end latency from the input data of the task to the output result. Since the input-output relationship between different functions affects the form of the objective function, three different objective functions are defined:

[0030] Serial functional dependency objective function:

[0031]

[0032] Among them, S is the task calculation delay under the serial function dependency, τ n Calculate the delay for the function, is the transmission delay of the function output result, n is the node number in the link, and N is the maximum number of nodes in the link;

[0033] Parallel functional dependency objective function:

[0034]

[0035] Where P is the task computation delay under the parallel row function dependency relationship, represents the data transmission delay of the task input function, ζ n is an intermediate variable, taking the value 0 or 1, and satisfying the constraints

[0036] Complex function dependency objective function:

[0037]

[0038] Where T is the task computation delay under complex function dependencies, ζ i is the intermediate variable, S i represents the serial data links decomposed from the complex functional dependency, and K represents the maximum number of serial data links.

[0039] Step 2.2: According to the target function, the terminal device generates an unloading strategy based on local computing resources, and uses a greedy strategy to determine the unloading position of the function according to the order of the function in the directed acyclic graph. At each decision, the location where the function is unloaded will be the node that contributes the most to the end-to-end delay of the task, that is, the minimum end-to-end delay of the function, until all functions are unloaded.

[0040] Since function computing offloading is a chain offloading method, based on this feature, a two-way greedy strategy, namely forward greedy and reverse greedy, is adopted in the algorithm design. Forward greedy starts from the function input data and unloads to the tail of the function dependency relationship in sequence, while reverse greedy starts from the tail of the function dependency relationship to the head. The forward and reverse directions converge in the middle of the function dependency relationship, and finally generate a complete offloading strategy.

[0041] Step 3: Find the multi-input and multi-output function dependencies in the offloaded functions, and use the data coordination mechanism to enhance data transmission capabilities, reduce data latency, and also reduce task computing latency;

[0042] Among them, the specific method of finding multi-input and multi-output function dependencies in the uninstalled function and using the data coordination mechanism to enhance the data transmission capability is as follows:

[0043] Step 3.1: Group the nodes. Figure 2 As shown, the data source node queues and numbers the requests in the order of their arrival time, and allocates bandwidth resources based on the bandwidth and node number of the data source node and the actual bandwidth demand. When the bandwidth of the data source node is allocated, the subsequent nodes that cannot obtain data will be mounted on the previous nodes within the tolerance delay range of the data, forming multiple data forwarding links, in which the data forwarding between each group does not affect each other.

[0044] The number of nodes in a group is calculated as follows:

[0045]

[0046] Among them, i and j represent the i-th and j-th edge computing nodes, is the packet delay from the i-th edge computing node to the j-th edge computing node. is the forwarding delay of the ith edge node, τ' is the tolerable delay, L k Represents a collection of nodes.

[0047] Step 3.2: Calculate the priority and sort the data forwarding order according to the priority. When grouping, the data forwarding order between nodes is uncertain. Different data forwarding strategies have a huge impact on data latency. Therefore, it is necessary to calculate the priority of the nodes in the group and sort the data forwarding order according to the priority.

[0048] like Figure 3 As shown, the priority calculation method consists of two steps: calculating the priority between nodes and calculating the priority within a group.

[0049] The priority calculation method between nodes is:

[0050]

[0051] Defined as

[0052]

[0053] in, is the optimal priority value between node i and node, represents the kth data link, k represents the number of the data link, γ represents the maximum value of the data link, α i,j represents the priority between nodes, D represents the amount of data, and B k,c represents the bandwidth from the node to the data source, B i,j Indicates the bandwidth between nodes;

[0054] The priority calculation method within a group is:

[0055]

[0056] Where P i represents the priority of node i in the link, α i,j Indicates the priority between node i and node j, and the node number in the n link.

[0057] Step 3.3: Data transmission. Data is transmitted according to the arrangement generated in step 3.2, and forwarded node by node until the last node. During the data transmission process, when the first node receives all the data, its bandwidth resources become idle. The idle bandwidth of the first node and the cloud computing node can be used to coordinate the data transmission to the subsequent nodes. The above process is repeated until the last node obtains all the data.

[0058] like Figure 4 As shown, the calculation method of the amount of data allocated to each node during data transmission is as follows:

[0059]

[0060] Among them, t ζ , D is the amount of data output by the function, B is the bandwidth, B' k represents the minimum bandwidth in the kth data link, B ζ＝1,C represents the bandwidth from node 1 to node C, B ρ,ζ represents the bandwidth from node ρ to node ζ, B ζ,C represents the bandwidth from node ζ to node C, ζ,ρ represents the ζ,ρth nodes, ζ,ρ represent nodes, and k represents the number of the data link.

[0061] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations on the present invention. A person skilled in the art may change, modify, replace and modify the above embodiments within the scope of the present invention without departing from the principles and intent of the present invention.

Claims

1. A cloud-edge-end function computing offloading method capable of data collaborative transmission, characterized in that: Combining the data collaborative transmission mechanism with function computing offloading includes the following steps: Step 1: In the cloud-edge-end collaboration framework, end users use the function computing mode to offload tasks to the cloud-edge collaborative network; Step 2: Determine the optimization objective function, calculate the offloading location of each function, determine the nodes involved in the task calculation and transmit the original data to the first function; Step 3: Find the multi-input and multi-output function dependencies in the offloaded functions, and use the data coordination mechanism to enhance data transmission capabilities, reduce data latency, and also reduce the computing latency of the task.

2. A cloud-edge-end function computing offloading method capable of data collaborative transmission according to claim 1, characterized in that: The specific steps of the function calculation mode in step 1 and determining the input-output relationship between functions and using a directed acyclic graph to represent the function dependency relationship within the task are as follows: Step 1.1: Deploy the FaaS platform on each node in the cloud-edge network and build a function library; Step 1.2: Obtain network parameters and obtain information about each node and task in the cloud-edge network through data interaction; Step 1.3: Construct a directed acyclic graph based on the decomposition relationship of the tasks to clarify the connection relationship between functions.

3. A cloud-edge-end function computing offloading method capable of data collaborative transmission according to claim 1, characterized in that: In step 2, the optimization objective function is determined, and the specific method for calculating the unloading position of each function is as follows: Step 2.1: Establish optimization objective function; Step 2.2: According to the target function, the terminal device generates an unloading strategy based on local computing resources, and uses a greedy strategy to determine the unloading position of the function according to the order of the function in the directed acyclic graph.

4. A cloud-edge-end function computing offloading method capable of data collaborative transmission according to claim 3, characterized in that: In step 2.1, since the input-output relationship between different functions will affect the form of the objective function, three different objective functions are established: Serial functional dependency objective function: Among them, S is the task calculation delay under the serial function dependency, τ n is the function calculation delay, ζ n is the transmission delay of the function output result, n is the node number in the link, and N is the maximum number of nodes in the link; Parallel functional dependency objective function: Where P is the task computation delay under the parallel function dependency relationship, represents the data transmission delay of the task input function, ζ n is an intermediate variable; Complex function dependency objective function: Where T is the task computation delay under complex function dependencies, ζ i is the intermediate variable, S i represents the serial data links decomposed from the complex functional dependency, and K represents the maximum number of serial data links.

5. A cloud-edge-end function computing offloading method capable of data collaborative transmission according to claim 3, characterized in that: The greedy strategy in step 2.2 is a bidirectional greedy strategy, namely forward greedy and reverse greedy; forward greedy starts from the function input data and unloads to the tail of the function dependency relationship in sequence, while reverse greedy starts from the tail of the function dependency relationship to the head. The forward and reverse directions converge in the middle of the function dependency relationship, and finally generate a complete unloading strategy.

6. A cloud-edge-end function computing offloading method capable of data collaborative transmission according to claim 1, characterized in that: The specific steps of step 3 are as follows: Step 3.1: Group the nodes; the data source node queues and numbers the requests in the order in which they arrive, and allocates bandwidth resources based on the bandwidth and node number of the data source node and the actual bandwidth demand. When the bandwidth of the data source node is allocated, the subsequent nodes that cannot obtain data will be mounted on the previous nodes within the tolerance delay range of the data, forming multiple data forwarding links; Step 3.2: Calculate the priority and sort the data forwarding order according to the priority; Step 3.3: Data transmission; forwarding node by node until the last node. During the data transmission process, when the first node receives all the data, its bandwidth resources become idle. The idle bandwidth of the first node and the cloud computing node is used to coordinately transmit the data to the subsequent nodes. The above process is repeated until the last node obtains all the data.

7. A cloud-edge-terminal function computing offloading method capable of data collaborative transmission according to claim 6, characterized in that: In step 3.1, the number of nodes in a group is calculated as follows: Among them, i and j represent the i-th and j-th edge computing nodes, is the packet delay from the ith edge computing node to the jth edge computing node, is the forwarding delay of the ith edge node, τ' is the tolerable delay, L k Represents a collection of nodes.

8. A cloud-edge-end function computing offloading method capable of data collaborative transmission according to claim 6, characterized in that: In step 3.2, the priority calculation method is a two-step process: calculating the priority between nodes and calculating the priority within a group; The priority calculation method between nodes is: Defined as: in, is the optimal priority value between node i and node α i,j represents the priority between nodes, D represents the amount of data output by the function, and B k,C represents the bandwidth from the node to the data source, B i,j Indicates the bandwidth between nodes; represents the kth data link, k represents the number of the data link, and γ represents the maximum value of the data link; The priority calculation method within a group is: Where P i represents the priority of node i in the link, α i,j Indicates the priority between node i and node j, and the node number in the n link.

9. A cloud-edge-end function computing offloading method capable of data collaborative transmission according to claim 6, characterized in that: In step 3.3, the calculation method for the amount of data allocated to each node during data transmission is as follows: Among them, t ζ , D is the amount of data, B' k represents the minimum bandwidth in the kth data link, B ζ＝1,c represents the bandwidth from node 1 to node c, B ρ,ζ represents the bandwidth from node ρ to node ζ, B ζ,C represents the bandwidth from node ζ to node C, ζ and ρ represent nodes, and k represents the number of the data link.

Images