Object Storage, Persistent Memory, and Data Infrastructure for HPC Materials Informatics

TL;DR

This paper discusses how object storage, persistent memory, and data infrastructure influence materials informatics, especially in the context of HPC and emerging storage-class memory for computational discovery of phase-change materials.

Contribution

It provides a conceptual analysis of data infrastructure choices in materials informatics, focusing on object storage, SCM, and non-von Neumann computing implications.

Findings

Infrastructure choices impact materials discovery workflows.

Emerging SCM hardware influences data management strategies.

Case study on phase-change materials illustrates practical implications.

Abstract

Speculation is provided on how infrastructure choices fit into the materials data ecosystem. Special attention is paid to object storage, the Intel DAOS API, storage-class memory (SCM), and the prospect of non-von Neumann computing. Lastly, the hypothesized implications of data infrastructure choices on a sample materials informatics problem is discussed: computational materials discovery of phase-change materials with properties tailored for phase-change memory (PCM). The motivation for selecting PCM as a sample materials informatics case study comes from its relevance to emerging SCM hardware.