Dynamic And Online Task Scheduling Based On Virtual Computing Group On Many-core Architecture

With the development of CPU, many-core architecture is the keypoint of the future development. How to enhance the running efficiency ofthe many-core system, realize the task scheduling heuristics properly andimprove the performance of CPUs are the hotspots of study.This paper focuses on the many-core processor which integrates tensto hundreds simple cores in a chip. In order to address the problems of taskscheduling and dynamic resource requirement for different tasks andphased resource requirement for the same task, the task model and thedynamic and online task scheduling and resources management heuristicsare proposed. These heuristics can reduce the costs of communication andimprove resources utilization and enhance throughput of the many-coreprocessors.In this paper, the main contributions can be divided into three parts,1)Considered the costs of communication and task phases, we proposedmapping heuristics based on the Virtual Computing Group.2) The resourcemanagement proposed can reduce the complexity of core distribution,compared to the current managements.3) Hundreds of tasks are createdrandomly and divided into several phases using our task model. Then a lotof simulations are done in our platform GEM5.According to the simulation result, our mapping algorithm canenhance the system throughput by18%and PE utilization by23%.