Unleashing In-network Computing on Scientific Workloads

TL;DR

This paper investigates how in-network computing can be adapted to high-performance scientific workloads by introducing NSinC, a closed-loop architecture that provides runtime feedback to optimize acceleration of scientific applications.

Contribution

It proposes NSinC, a novel architecture that enables in-network computing to effectively support scientific workloads through runtime feedback mechanisms.

Findings

Identifies challenges of applying in-network computing to scientific workloads.

Proposes a preliminary design for a closed-loop in-network acceleration system.

Highlights potential benefits of in-network computing for HPC applications.

Abstract

Many recent efforts have shown that in-network computing can benefit various datacenter applications. In this paper, we explore a relatively less-explored domain which we argue can benefit from in-network computing: scientific workloads in high-performance computing. By analyzing canonical examples of HPC applications, we observe unique opportunities and challenges for exploiting in-network computing to accelerate scientific workloads. In particular, we find that the dynamic and demanding nature of scientific workloads is the major obstacle to the adoption of in-network approaches which are mostly open-loop and lack runtime feedback. In this paper, we present NSinC (Network-accelerated ScIeNtific Computing), an architecture for fully unleashing the potential benefits of in-network computing for scientific workloads by providing closed-loop runtime feedback to in-network acceleration services. We outline key challenges in realizing this vision and a preliminary design to enable acceleration for scientific applications.