Cluster many-core processor system and directed acyclic graph task dynamic cluster selection method

Abstract

The invention discloses a cluster many-core processor system and a directed acyclic graph task dynamic cluster selection method, and belongs to the technical field of high-performance calculation and task scheduling, the system comprises a directed acyclic graph task input module, a cluster resource monitor, a scheduler, a task distributor and a task execution unit; inputting predecessor task node information containing task nodes to be executed and a core number required by task execution to the acyclic graph task; the cluster resource monitor monitors the number of idle cores of each cluster in real time according to the state information of each core in the cluster, and inputs the information into the scheduler for task scheduling; core decision logic for task cluster selection is integrated in the scheduler, and a proper computing cluster is dynamically selected for a task to be executed. According to the method, the task scheduling efficiency can be optimized, the communication overhead can be reduced and the utilization rate of system resources can be improved aiming at the dynamic cluster selection of the directed acyclic graph task model in a cluster many-core processor environment.

Images