 |
||
|
||
Home > TERATEC FORUM > Workshops > Workshop 6
Workshops
Thursday June 24
Workshop 06 - 16:00 to 18:00
New Storage Paradigms to Tackle Exascale Challenges
Chaired by Philippe Deniel, chef du Laboratoire architecture du stockage et des systèmes, CEA DAM
SHAMan: a flexible framework for auto-tuning HPC systems
By Sophie Robert, Doctorante, Atos,
Soraya Zertal, Maitre de Conférence Hors Classe, UVSQ
et Philippe Couvée, Senior Software Architect, ATOS
Modern computer components, both hardware and software, come with many tunable parameters. Making sure that the parametrization of these components is optimal is crucial for maximizing the performance of computer systems. This is especially important for HPC systems, as many components of a cluster can be parametrized, such as Message Passing Interface libraries (MPI), storage bays or I/O (Input/Output) accelerators.
Auto-tuning methods relying on black-box optimization have delivered good results for finding the optimal parametrization of complex computer systems. We propose a new optimization framework, called the Smart HPC MANager. It provides an out-of-the-box Web application to perform black-box auto-tuning of computer components running on a distributed system for a user application.
Currently, this framework integrates three state-of-the-art heuristics, resampling strategies to deal with the noise due to resource sharing, and pruning strategies to speed-up the convergence process.
We validated with a significant improvement percentage, three use-cases of SHAMan by tuning a software I/O accelerators (Small Read Optimizer), a hardware I/O accelerators (Smart Burst Buffer) and MPI libraries.
 |
Biography: Sophie Robert is a third year PhD student at the University of Versailles (UPSaclay-UVSQ) in France, where she is supervised by Dr Soraya Zertal. She is currently working on the auto-optimization of complex and noisy HPC systems, and focuses especially on the tuning of I/O accelerators. Her research grant is partially funded by the Atos company where she is supervised by Philippe Couvée. Her original background is in applied mathematics, as she graduated with a master in applied mathematics, from the University of Auvergne, France in 2017 and a post-master degree in Data Science from ENSIMAG, France in 2018. |
|---|---|
 |
Biography: Soraya Zertal is an associate professor with accreditation to supervise research at LiPaRAD lab. Part of UPSaclay-UVSQ university. She has a Master and a PhD in computer science from the university of Versailles, France. Her main research interests are related to: I/O and storage systems, object storage, Storage devices technologies (HDD, SSD, NVRAM), data placement strategies, storage systems robustness, modeling and simulation for performance evaluation and tuning applied to IO accelerators. She worked on different national/international, academic and industrial projects and supervised several Master and PhD students in these topics. |
 |
Biography: Philippe Couvée is working in HPC R&D since more than 20 years. He is leading a team of 17 researchers and engineers developing products that facilitates data access from large supercomputers. Its recent focus is on data centric solutions that combine cache and acceleration technics, in-situ and in-transit processing with advanced instrumentation and data analytics. He is also teaching computer architecture and system programming at CNAM |
Register now and get your badge here
- TERATEC Forum is strictly reserved for professionals.
- Participation to exhibition, conferences and workshops is free (subject to seats available)
- On line registration is obligatory.