Joint Reducer Placement and Coflow Bandwidth Scheduling for Computing Clusters

Reducing Coflow Completion Time (CCT) has a significant impact on application performance in data-parallel frameworks. Most existing works assume that the endpoints of constituent flows in each coflow are predetermined. We argue that CCT can be further optimized by treating flows' destinations as an additional optimization dimension via reducer placement. In this article, we propose and implement RPC, a joint online Reducer Placement and Coflow bandwidth scheduling framework, to minimize the average CCT in cloud clusters. We first develop a 2-approximation algorithm to minimize the CCT of a single coflow, and then schedule all the coflows following the Shortest Remaining Time First (SRTF) principle. We use real testbed experiments and extensive large-scale simulations to demonstrate that RPC can reduce the average CCT by 64.98% compared with the state-of-the-art technologies.