A new task mapping technique for communication-aware scheduling strategies

Abstract

Clusters have become a very cost-effective platform for high-performance computing. In these systems, the trend is towards the interconnection network becoming the system bottleneck. Therefore, in the future, scheduling strategies will have to take into account the communication requirements of the applications and the communication bandwidth that the network can offer. One of the key issues in these strategies is the task mapping technique used when the network becomes the system bottleneck. In this paper, we propose an enhanced version of a previously proposed mapping technique that takes into account not only the existing network resources, but also the traffic generated by the applications. Also, we evaluate the mapping technique using real MPI application traces with timestamps. Evaluation results show that the rise of the new mapping technique fully exploits the available network bandwidth, improving load balancing and increasing the throughput that can be delivered by the network.