Multi-coflow scheduling method based on graph neural network deep reinforcement learning

Abstract

The invention discloses a multi-coflow scheduling method based on graph neural network deep reinforcement learning. A multi-coflow scheduling model is established based on a deep reinforcement learning framework; a cascaded graph neural network and strategy network is adopted as a deep reinforcement learning agent, wherein the workflow DAG feature extraction is completed by a graph neural network, so that the model can process the workflow DAG with different numbers and connection modes of nodes, and the generalization ability of the model under the unpredictable input DAG is effectively improved; by introducing the strategy converter, a fine-grained coflow scheduling strategy can be generated according to the scheduling priority list, the efficiency of the scheduling process is improved,and the completion time of workflow is effectively shortened.

Description

technical field [0001] The invention belongs to the technical field of computer networks, and in particular relates to a multi-coflow scheduling method based on graph neural network deep reinforcement learning. Background technique [0002] Cluster computing is a method that has been widely adopted in recent years because it can achieve high parallel computing performance at a relatively low capital expenditure as data processing requirements increase in scenarios such as big data and cloud computing. [0003] Many cluster computing applications (such as Spark and MapReduce) are based on a multi-tasking model, where a computing task consists of several consecutive stages and communication stages. Each stage of execution calculation can only start when the intermediate data of the previous stage has been completely sent. The working process of a task can be modeled as a workflow directed acyclic graph (Directed Acyclic Graph, DAG) denoted as G=(V,E), where V is the node set ...

AI Technical Summary

Problems solved by technology

[0005] The coflow scheduling problem has always been a challenging problem in data center networking (DCN), because coflow has various characteristics that complicate the scheduling of coflow

[0006] Some existing technologies only focus on minimizing the transmission time of a communication stage. In the scheduling process, the communication requirements of the workflow are not considered. This kind of workflow-independent scheduling method does not consider the execution of the work DAG in the coflow scheduling process. order, so the optimal workflow completion time cannot be calculated; some existing technologies use heuristic algorithms to simplify the problem, and such methods can only guarantee the optimal solution close to the NP-hard problem

Therefore, there is still a lot of room for improvement in solving the workflow completion time

Images