Journées Thèmes Emergents - HPC File Systems: From Cluster to Grid
http://www.irisa.fr/paris/alebre/JTE-HPC-File-Systems.html
October 3-4, 2007 IRISA, Rennes
co-organized by the PARIS project-team and the french division of ACM SIGOPS
October 3, 2007 Wednesday (Room Métivier)
13:45
Greeting
14:00
High Performance Lustre Storage Architecture at CEA/DIF
Stephane Thiell, CEA/France
14:45
Toward An Integrated Cluster File System
Adrien Lebre, PostDoc, IRISA
15:30
Coffee break
15:45
LHC Data Management On A Global Grid
Fabio Hernandez, IN2P3/CNRS Computing Center, Lyon, France
16:30
The Visage project - Storage Virtualization At Grid Level
Francois Thiebolt, Salam Traboulsi, Phd Candidate, IRIT, Toulouse, France
This presentation will mainly focus on the VisageFS aspects. Further details
17:15
The Gedeon Project - Data, Meta-data and Databases
Yves Denneulin, Associate Professor, LIG, Grenoble, France
Overview of the Gedeon solution, cache strategies, API (Brief demonstration on bio-informatic data)Further details
18:00
XtreemFS - an Object-Based File System for Large-Scale Federated IT Infrastructures
Jan Stander, ZIB, Berlin, Germany
XtreemFS is a POSIX-compliant object-based file system which supports replication of file data and metadata over WAN connections spanning multiple administration domains, as well as data striping and metadata partitioning. In our talk, we will describe the architectural requirements for extending object-based storage to federated installations, discuss the issues that arise when replicating file data and metadata over WAN connections, and give an overview of the XtreemFS implementation and protocols. Further details
18:45
20:30 Dinner in city center
October 4, 2007 Thursday (Room Métivier)
09:00
GFarm: a Global File System
Osamu Tatebe, Associate Professor, University of Tsukuba, Japan
Gfarm Grid file system has been developed for facilitating reliable file sharing and high-performance data computing in a Grid across administrative domains. It is a scalable virtual file system federating local file systems of cluster nodes. Efficient utilization of local file system of compute nodes is a key technology for scalable parallel and distributed data computing. This talk describes the design and implementation of a secure, robust, scalable and high-performance Gfarm file system. Also, this talk includes how we won the HPC Storage Challenge at SC06.Further details
09:45
Pastis, a peer-to-peer file system for persistant large-scale storage
Pierre Sens, Professor, Université Paris VI, France
Further details
10:30
Coffee break
10:45
Industrial Session: Futur Directions for HPC File systems
12:30
Open Discussions ....