Home // FUTURE COMPUTING 2021, The Thirteenth International Conference on Future Computational Technologies and Applications // View article

FOSDA: A Hybrid Disaggregated HPC Architecture based on Distributed Nanoseconds Optical Switches

Authors:
Xiaotao Guo
Xuwei Xue
Bitao Pan
Fulong Yan
Georgios Exarchakos
Nicola Calabretta

Keywords: HPC network; disaggregated architecture; fast optical switch.

Abstract:
Aiming at solving the issues of low resource utilization and high operational cost in current node-centric High Performance Computing (HPC) architecture, we present and investigate a novel hybrid disaggregated HPC architecture based on nano-seconds fast optical switch (FOSDA). The fast optical switch connects Central Processing Unit (CPU) and memory nodes for the communication of high bandwidth and low latency, while storage nodes are interconnected by electrical packet switches. The performance of the FOSDA in terms of workload acceptance rate, resource utilization, power consumption and capital/operational cost is numerically assessed and compared with current node-centric HPC architectures. Compared with a node-centric HPC architecture of 320 nodes, FOSDA performs 36.6% higher CPU, and 21.5% higher memory resource utilization with 45.5% less active hardware resource, as well as 46.8% less power consumption. When scaling the HPC network to 2304 nodes, FOSDA also achieves 33.6% higher CPU and 48.5% higher memory utilization while saving 50.4% power consumption. In the cost analysis, FOSDA decreases operational cost by 46.7% with only 19.8% more capital cost.

Pages: 1 to 7

Copyright: Copyright (c) IARIA, 2021

Publication date: April 18, 2021

Published in: conference

ISSN: 2308-3735

ISBN: 978-1-61208-846-4

Location: Porto, Portugal

Dates: from April 18, 2021 to April 22, 2021