An Approach for Parallel Scheduling for Agile Co Scheduling

Co scheduling deals with minimizing the communication waiting time of parallel processes by identifying the processes in need of Co scheduling through gathering and analyzing implicit runtime information, basically communication events. Synchronization needs of some applications can suffer severe performance penalties if their processes are not all co scheduled to run together. Two common approaches to co scheduling jobs are batch scheduling, where in nodes are dedicated for the duration of the run, and gang scheduling, wherein time slicing is coordinated across processors. Both work well when jobs are load-balanced and make use of the entire parallel machine. ACS (Agile Co Scheduling) addresses this problem by monitoring each job’s computation granularity and communication pattern and scheduling jobs based on their synchronization and load-balancing requirements. ACS has been fully implemented on top of the STORM (Scalable TOol for Resource Management (Scheduling Algorithm)) resource manager on batch, gang, and implicit co scheduling algorithms. This paper describes in detail the implementation of ACS and Its performance evaluation with dynamic workloads. Our results are positive in that we demonstrate efficiency and portability over a range of parallel Scheduling and show that the FCFS model is useful for predicting performance trends and leads to reduced utilization and suboptimal performance. In order to solve these problems, a new technique, named Cooperating Agile Co Scheduling (CCS), is presented in this work.