J.-L. Yu and S.-R. Maeng (Korea)
Communication-driven coscheduling, Priority boost con flict, Resource contention, Runtime rescheduling latency, Performance evaluation, Heterogeneous clusters
Communication-driven coscheduling (CDC) has been demonstrated to be an effective technique to enhance the performance of parallel applications in multi-programmed clusters. However, existing CDC schemes do not incorpo rate any steps to adequately deal with priority boost con flicts, leading to significantly degraded performance. To address this problem, in this paper, we propose a flexible CDC scheme that regulates the scheduling se quence of conflicting processes based on the rescheduling latency of their correspondents in remote nodes. We also demonstrate the implementation details of the proposed scheme built on top of an unified coscheduling frame work which can be used to implement any coscheduling schemes with minimal effort. Using this framework, we ex haustively explore the impact of various system parameters and/or job characteristics on the performance of all consid ered CDC schemes on a heterogeneous cluster testbed. The results show that the proposed approach outperforms exist ing schemes (by up to 36.6% in average job response time), reducing both boost conflict ratio and message delay.
Important Links:
Go Back
