Ahmad Hossein Yazdani

I am Ahmad Hossein Yazdani, an ML systems researcher with a PhD in Computer Science from Virginia Tech. My research spans systems for ML and ML for systems, focusing on memory and I/O bottlenecks in large-scale AI and HPC workloads, including distributed LLM training and intelligent I/O optimization. Through research internships at Oak Ridge National Laboratory and Lawrence Berkeley National Laboratory/NERSC, I have developed ML- and agent-based techniques for diagnosing and improving performance across GPU clusters and parallel storage systems.

My research interests include distributed systems optimization, resource management, cloud computing, and high-performance computing (HPC) for AI/ML workloads.

news

Jul 22, 2026	Today I successfully defended my PhD 🎉🎉 Many thanks to my advisor Dr Ali Butt, my committee members, my collaborators and my labmates over the past 6 years for their support.
Jan 23, 2026	I’m excited to announce that I kicked off my internship at Oak Ridge National Laboratory on this day.
Oct 02, 2024	I’m thrilled to announce that our paper titled “User-based I/O Profiling for Leadership Scale HPC Workloads” has been accepted to ICDCN25 to be held in Haydarabad, India January 4th to 7th 2025.
Jun 12, 2024	I’m honored to announce that I started off another internship at NERSC, Lawrence Berkeley National Laboratory in this day.
May 26, 2024	I had the honor to present a poster about our project on I/O interference characterization with the Berkeley Lab (LBNL) at the IPDPS PhD forum.

selected publications

ICDCN’25
User-based I/O Profiling for Leadership Scale HPC Workloads

Ahmad Hossein Yazdani, Arnab K. Paul, Ahmad Maroof Karimi, and 2 more authors

In Proceedings of the 26th International Conference on Distributed Computing and Networking, Heydarabad, India, Jan 2025

Abs DOI Bib

I/O constitutes a significant portion of most of the application run-time. Spawning many such applications concurrently on an HPC system leads to severe I/O contention. Thus, understanding and subsequently reducing I/O contention induced by such multi-tenancy is critical for the efficient and reliable performance of the HPC system. In this study, we demonstrate that an application’s performance is influenced by the command line arguments passed to the job submission. We model an application’s I/O behavior based on two factors: past I/O behavior within a time window and user-configured I/O settings via command-line arguments. We conclude that I/O patterns for well-known HPC applications like E3SM and LAMMP are predictable, with an average uncertainty below 0.25 (A probability of 80%) and near zero (A probability of 100%) within a day. However, I/O pattern variance increases as the study time window lengthens. Additionally, we show that for 38 users and at least 50 applications constituting approximately 93000 job submissions, there is a high correlation between a submitted command line and the past command lines made within 1 to 10 days submitted by the user. We claim the length of this time window is unique per user.
@inproceedings{10.1145/3700838.3700865, author = {Yazdani, Ahmad Hossein and Paul, Arnab K. and Karimi, Ahmad Maroof and Wang, Feiyi and Butt, Ali}, title = {User-based I/O Profiling for Leadership Scale HPC Workloads}, year = {2025}, isbn = {9798400710629}, publisher = {Association for Computing Machinery}, address = {New York, NY, USA}, url = {https://doi.org/10.1145/3700838.3700865}, doi = {10.1145/3700838.3700865}, booktitle = {Proceedings of the 26th International Conference on Distributed Computing and Networking}, pages = {181–190}, numpages = {10}, keywords = {High Performance Computing, I/O characterization, I/O scheduler, Darshan, I/O profiling}, location = {Heydarabad, India}, series = {ICDCN '25}, month = jan, }

FAST’23

SHADE: Enable Fundamental Cacheability for Distributed Deep Learning Training

Redwan Ibne Seraj Khan, Ahmad Hossein Yazdani, Yuqi Fu, and 5 more authors

In 21st USENIX Conference on File and Storage Technologies (FAST 23), Feb 2023

Bib PDF Code

@inproceedings{285774,
  author = {Khan, Redwan Ibne Seraj and Yazdani, Ahmad Hossein and Fu, Yuqi and Paul, Arnab K. and Ji, Bo and Jian, Xun and Cheng, Yue and Butt, Ali R.},
  title = {{SHADE}: Enable Fundamental Cacheability for Distributed Deep Learning Training},
  booktitle = {21st USENIX Conference on File and Storage Technologies (FAST 23)},
  year = {2023},
  isbn = {978-1-939133-32-8},
  address = {Santa Clara, CA},
  pages = {135--152},
  url = {https://www.usenix.org/conference/fast23/presentation/khan},
  publisher = {USENIX Association},
  month = feb,
}