Composability of Cloud Accelerators in Virtual World Simulations

Abstract

Immersive and interactive experiences offered by virtual world simulations (VWS) are becoming increasingly indispensable in today's digital era, particularly in domains such as gaming, engineering, and education. Cloud computing provides a compelling solution for VWS, with its scalability, costeffectiveness, and accessibility, combined with efficient processing capabilities of hardware accelerators such as FPGAs and GPUs. This paper presents APEIRON, an in-progress distributed library built on the Ray framework, that aims to facilitate the effective composition of hardware accelerators on public clouds and emerging disaggregated platforms. Additionally, this work introduces a novel composability metric, analyzing the intricate relationship between resource utilization and workload performance. This study focuses on theoretical composability analyses, laying groundwork for optimized utilization of hardware resources to enhance efficiency, thereby potentially reducing operational costs.