Toward an Enterprise-ready Composable Infrastructure as a Service

A composable infrastructure is defined as resources such as compute, storage, accelerators, and networking, that are shared in a pool and that can be grouped in different configurations to varying meet business requirements. In this paper, we explore using the flexible environment of a Composable In...

Full description

Saved in:
Bibliographic Details
Main Authors: Herger, Lorraine M., El Maghraoui, Kaoutar, Chung, I-Hsin, Choudary, Chekuri, Tran, Kim, Deshane, Todd
Format: Conference Proceeding
Language:English
Subjects:
Benchmark testing
Cloud computing
Composable
Composable Infrastructure as a Service
Composable System Service
Costs
Dis-aggregated Systems
IaaS
Resource Disaggregation
Resource Management
Software architecture
Software design
Technological innovation
Topology
Online Access:Request full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:A composable infrastructure is defined as resources such as compute, storage, accelerators, and networking, that are shared in a pool and that can be grouped in different configurations to varying meet business requirements. In this paper, we explore using the flexible environment of a Composable Infrastructure as a Service (CIaaS), to innovate in system and software design. Early experimentation provides insights into a hardware or software architecture to identify key performance bottlenecks, improve performance, reliability, cost, etc., this avoiding costly redesign. Systems and software evolve quickly, with innovations such as System on a Chip (SOC), and new accelerator types. Additionally, High performance computing (HPC) and AI workloads put pressure on designers to develop quickly, yet meet business goals. This paper details the design of an enterprise CIaaS that can be used for early experimentation, via a service model. Our experimental evaluations give insights into how the system works, evaluates the impact of various resource aggregations and reconfigurations on representative benchmarks, and show how these results can be used to inform the design of new systems. The use cases demonstrate using the CIaaS as an efficient, cost-effective approach to resource provisioning in data centers. In summary, our contributions are centered on 1. capability to dynamically dis-aggregate and compose resources, and 2. the enterprise readiness of the system, which allows it to be placed safely on an Extranet.
ISSN:2474-2473
DOI:10.1109/SCC53864.2021.00024