Noah Lewis

Noah Lewis

Auto-generated profile, based on publicly available information

Noah Lewis

Computer Science, The Ohio State University

Research Focus & Metrics

Academic Career Length

2Years (2025-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
7 May 2025 in ACM Computing Surveys
Verified
Journal Article
Open access
NLNoah Lewis
Noah Lewis
JBJean Luca NMI Bez
Jean Luca NMI Bez
SBSurendra Byna
Surendra Byna
Growing interest in Artificial Intelligence (AI) has resulted in a surge in demand for faster methods of Machine Learning (ML) model training and inference. This demand for speed has prompted the use of high performance computing (HPC) systems that excel in managing distributed workloads. Because data is the main fuel for AI applications, the performance of the storage and I/O subsystem of HPC systems is critical. In the past, HPC applications accessed large portions of data written by simulations or experiments or ingested data for visualizations or analysis tasks. ML workloads perform small reads spread across a large number of random files. This shift of I/O access patterns poses several challenges to modern parallel storage systems. In this article, we survey I/O in ML applications on HPC systems, and target literature within a 6-year time window from 2019 to 2024. We define the scope of the survey, provide an overview of the common phases of ML, review available profilers and benchmarks, examine the I/O patterns encountered during offline data preparation, training, and inference, and explore I/O optimizations utilized in modern ML frameworks and proposed in recent literature. Lastly, we seek to expose research gaps that could spawn further R&D.

14 November 2026 in ACM Computing Surveys
Verified
Journal Article
Open access
ASAmir Salimiparsa
Amir Salimiparsa
NLNoah Lewis
Noah Lewis
JBJean Luca NMI Bez
Jean Luca NMI Bez
SBSurendra Byna
Surendra Byna
High-performance computing (HPC) storage and I/O software stacks have been challenging to configure due to multiple software layers, interdependencies among those layers, and the number of parameters to tune within each layer. The storage stack has become increasingly complex with the ever-growing sets of new software and hardware. In addition, there is a shift in HPC applications from modeling and simulations (mod-sim) to AI training and inference. While several efforts focused on tuning I/O parameters of write-heavy simulation workloads, configuring I/O for random and read-heavy AI applications has still been in progress. Although the HPC storage stack could be reconfigured to meet these diverse write and read requirements, finding optimal I/O configurations for even a single application is a challenging task. This survey proposes a taxonomy of HPC storage and I/O system configurations and reconfigurability for bringing a consensus to the I/O parameters that are tunable by clients (i.e., users and applications). In this paper, we survey I/O tuning and optimization, harnessing the HPC I/O system configuration knowledge from the past decade. This paper can serve as a primary source of information for HPC I/O researchers and developers to design and optimize novel tuning strategies that improve storage and I/O system reconfigurability and performance.