Cloud computing and EDA (Electronic Design Automation) : cost optimization for the EDA jobs with object storage

Abstract

Cloud computing with its ability to provide on-demand and optimized computing resources is emerging to disrupt the High-Performance Computing (HPC) markets. On the semiconductor domain, the Electronic Design Automation (EDA) has been an essential prerequisite, enabling the design of complex integrated circuits. With the increasing complexity of System on the Chip (SoC) designs, coupled with the need for faster time-to-market has led to a growing demand for scalable cost-effective EDA solutions. The emergence of cloud computing has provided a potential solution to these challenges, offering the ability to scale EDA workflows on-demand and pay- per-usage concept. The thesis explores the avenues in using JuiceFS, the open source, high-performance, POSIX compatible, distributed file system designed for the native cloud applications to deploy EDA workloads in the cloud with the aid of object storage by minimizing the cost for storage. Furthermore, characteristics of EDA workloads are identified. The cost for storage (eg. High-performance Network File System (NFS), AWS FSx for Lustre and Azure NetApp) is a major contributor to the overall cost of the cloud and has been a limiting factor for most of the semiconductor companies, especially for the startups. The storage cost component can be minimized by enabling the file system with cheaper object storage. The tradeoff is the performance downgrade (Run time of the EDA tool) and it is observed that high- performance POSIX-compatible file system is required to minimize the tradeoff. An experiment on VCS* compile time is conducted to identify the tradeoffs using JuiceFS configured with the metadata server and object server. *Collaborating with Synopsys Inc., the VCS slowdown in the JuiceFS setup is analyzed with the Proof of Concept (POC) developed by Synopsys Inc. on top of JuiceFS.