CSR: Small:Conductor: A Run-Time System for Exascale Computing

University Of Arizona, Tucson AZ

Investigators

Abstract

One of the critical problems---if not the critical problem---in reaching the exascale computing goal by the end of the decade is the power problem. Exascale systems have a target power constraint of 20 megawatts even though today's petascale systems---which have performance at least two orders of magnitude below prospective exascale systems---generally consume around 5 megawatts. Hardware improvements alone will not bridge this gap. The PI is developing a run-time system called Conductor to address the power issue. The overall goal of Conductor is to produce near-optimal application performance under a prescribed power bound. Conductor carries this out by allocating power both between and within nodes. First, the PI is designing and implementing a new technique called power scheduling, which addresses the inter-node case. The second part of Conductor addresses the intra-node case with power gating, which allows powering off of individual components in a more fine-grain manner than is generally available in architectures today. Finally, the PI is investigating techniques to allow users to assist Conductor, through annotations, in achieving high performance in cases where power scheduling and power gating alone are not sufficient. The impact of the research described in this proposal is significant. Conductor is essential to achieving exascale performance on nontrivial applications, and it will help push towards the exascale goal. Achieving exascale performance is an important national priority and will impact many application domains. The PI is also integrating the research ideas into both undergraduate and graduate parallel and distributed computing courses.