Dynamic Scheduling Method Of EDA Scientific Workflows In Hybrid Computing Environments

With the increasing complexity of integrated circuits,electronic design automation(EDA)technology has entered a stage of rapid development.How to use EDA technology to design VLSI skillfully and efficiently has become the focus of many researchers.EDA scientific workflow refers to the computational task flow in the process of designing integrated circuits using EDA technology,including the combination of tasks in various stages of EDA design.Because the amount of computation in the process of designing an integrated circuit is constantly increasing,in order to ensure design efficiency,supercomputer is required to provide a large amount of computing resources.However,EDA tasks in multiple EDA scientific workflows submitted by multiple users require different operating environments.Single computing resource cannot support the operation of EDA tasks efficiently.A hybrid computing environment of cloud and high performance clusters is used to allocate resources reasonably.At the same time,commercial licenses for EDA tools are not only expensive but also have different types and quantities,which requires optimal matching of tasks and resources.In view of these characteristics and constraints,the existing multi-workflow scheduling algorithms are not suitable for EDA scientific workflow.In the hybrid computing environments,efficient and dynamic scheduling of EDA scientific workflows is of great help to the development of EDA industry,and is a scientific issue that needs to be solved urgently.In this thesis,a dynamic EDA Scientific Workflows Scheduling(DESWS)method is proposed for multi-user based on hybrid computing environment.The DESWS method defines the dynamic maximum waiting time based on the estimated execution time,required resources,and overall resources of the EDA task in the EDA scientific workflow,and uses the double-layer priority mechanism to determine the priority of task scheduling.By this means,user fairness is obtained and the quality of service is improved.In addition,in order to ensure that each EDA task is allocated the most reasonable resources and run it efficiently in the hybrid computing environment,an efficient hybrid resource allocation strategy is proposed based on task queues and maximizing resource utilization.Consider the resource types,sizes,and licenses required for EDA tasks,assign available resources to tasks based on task priority order and achieve reasonable matching of tasks and resources to optimize overall makespan.In the hybrid computing environment,the DESWS method is more efficient and reasonable for the EDA scientific workflow than the traditional scheduling method,and the balance between user fairness and overall makespan is obtained,which reduces overall makespan by up to 26.9%,and improves user fairness by 23.04%.