Contention-aware Communication-driven Co scheduling Strategy for Multi-Programmed Heterogeneous Clusters
Jung-Lok Yu, Heejung Byun · Jeongbo gwahaghoe nonmunji. si'seu'tem mich i'lon · 2011
Communication-driven coscheduling (CDC) has been demonstrated to be an effective technique in enhancing the performance of parallel applications in multi-programmed clusters. However, one major problem found in existing CDC schemes is that they do not incorporate any steps to attempt to properly handle priority boost conflicts, leading to significantly degraded performance. To address this problem, we propose a novel CDC scheme (called CC: Contention-aware Coscheduling) that regulates the scheduling sequence of conflicting processes based on the rescheduling latency of their correspondents in remote nodes. We also present an unified coscheduling framework which can be used to implement any co scheduling schemes with minimal effort, and implementation details of the proposed scheme. Using this framework, we exhaustively explore the impact of various system parameters and/or job characteristics on the performance of all considered CDC schemes on a heterogeneous cluster. The results show that the proposed approach outperforms existing schemes (by up to 36.6% in average job response time), reducing both boost conflict ratio and message delay.