[communities] GGF Proposal Submission

[ctjordan at sdsc.edu]

proposers_name: Chris Jordan 
 
affiliation: San Diego Supercomputer Center 

email: ctjordan at sdsc.edu 

proposed_title: A Wide-Area Parallel Filesystem for the TeraGrid: A Case Study 

session_type: Presentation (two presenters) 

proposed_duration: 60 mins 

target_audience: Users/Engineers 

num_attendees: 50-75 

abstract: TeraGrid deployed the first production, grid-enabled high performance parallel file system using 500 TB of disk and IBM General Parallel File System (GPFS). Challenges such as latency, which were overcome in the development of the deployment model are explained, with notes on the strategies taken in overcoming these challenges. The use of existing Grid technologies to enable UID-mapping between administrative domains is described in detail, along with the relationship of this work to evolving Grid standards work. The experience of pilot projects such as NVO, ENZO, and BIRN, and the production workflows enabled by the global filesystem are related. 

synopsis: Session Goals:
After the session is completed, attendees should understand the need for wide-area, Grid-enabled filesystems, and the particular approach to the problem taken in this project. Attendees should understand the relationship of the GPFS-WAN work to the evolution of Grid standards and existing Grid technologies. Attendees who are users should understand how a Grid file system can enable new workflows, and attendees who are engineers should understand the specific technical challenges inherent in large Grid-based file system deployment.

Outline:

Introduction to TeraGrid
Motivation - TeraGrid user needs
Brief Introduction to GPFS
Previous work on Grid file systems at SDSC
Security Requirements
Performance Requirements
Reliability Requirements

Initial Hardware Resources and Configuration
Cluster Authentication and Authorization
User Identification and UID-mapping
GSI-based UID-mapping

Special Considerations for Grid File System design
Metadata and Management servers
Data Replication Options
Final Hardware Configuration

Network Performance Issues
OS-Specific and GPFS Network tuning
Local and Remote performance benchmarks
Network Reliability Issues

Pilot User Projects - BIRN and NVO
Overview of BIRN/lddmm characteristics
Overview of NVO/2MASS characteristics
ENZO use of GPFS-WAN
BIRN use of GPFS-WAN
2MASS use of GPFS-WAN
Multi-site task workflows

GPFS-WAN and Grid Standards
Access mechanisms/Interface issues
Policies for Grid File Systems
GFS WG/RNS
OGSA Data WG
Challenges for Grid File System Standards

Summary - Achievements and Goals
GPFS-WAN as a Production Grid Resource
User Adventures
Future Work
 

tech_requirements: None 

prereq_participants: Basic knowledge of parallel file system concepts and the Globus Toolkit security layer will be helpful. 

advertise_suggestion: The work described is relevant to the OGSA-Data, GFS, GSM, and ByteIO working groups, at least, and each of these WG mailing lists should be informed of the presentation, should it be accepted. TeraGrid news items and mailing lists could also be used to inform interested individuals of the presentation.