Home // CLOUD COMPUTING 2022, The Thirteenth International Conference on Cloud Computing, GRIDs, and Virtualization // View article

Gestalt Computing: Hybrid Traditional HPC and Cloud Hardware and Software Support

Authors:
Jay Lofstead
Andrew Younge

Keywords: HPC, Cloud, hybrid computing, hybrid workloads, persistent memory, job scheduling

Abstract:
Traditional modeling and simulation-based HPC workloads demand a platform optimized for low latency node-to-node communication and a write intensive IO load. Cloud-based applications tend to not need the fast communication and are frequently heavily read dependent. Both of these workloads are immensely important and demand larger and larger machines to handle the demand. Unfortunately, budgets demand that large scale compute be a shared resource for both workloads in spite of this opposed design priority. Through the use of persistent memory (PMEM) devices on node, dynamically we can reconfigure the nodes to either support fast reading through using the on node-PMEM as a reading cache or as slow DRAM to reduce the time spent communicating with logical neighbor nodes. We outline a vision for a machine hardware and software architecture dynamically shifting to simultaneously support both workloads as demand dictates with minimal additional cost and complexity.

Pages: 23 to 30

Copyright: Copyright (c) The Government of Sandia National Laboratories, 2022. Used by permission to IARIA.

Publication date: April 24, 2022

Published in: conference

ISSN: 2308-4294

ISBN: 978-1-61208-948-5

Location: Barcelona, Spain

Dates: from April 24, 2022 to April 28, 2022