Unified Science Environment (USE)

This year NERSC took some major steps toward the goal of a Unified Science Environment with the installation and testing of prototype Grid services, the establishment of several Grid testbeds, and another award-winning demonstration of distributed computing and visualization in the SC2002 Bandwidth Challenge competition. All of NERSC’s major computing and storage systems are expected to become accessible via the DOE Science Grid in 2003 when all of the initial components of this infrastructure are placed in production.

Establishing a Grid Infrastructure

NERSC had originally planned to make its high performance computing systems accessible via the DOE Science Grid by 2004, but the target date was moved up a year when NERSC established a collaboration with IBM to deploy Grid capabilities on the SP and HPSS systems. The goal of this collaboration is to seamlessly integrate IBM’s system software with the Globus Grid software. The Globus Toolkit 2 underwent tests on NERSC’s development SP cluster during the past year, and some minor problems were identified for correction. DOE Science Grid access to Seaborg is expected by mid-2003. The Globus Security Infrastructure has been integrated with most HPSS tools, and single-stream file transfers to and from HPSS via the Grid should be available on a production basis in early 2003, with parallel file transfers coming a year later. NERSC is working with Argonne National Laboratory to develop a full-featured, Grid-enabled HPSS.

NERSC has been collaborating with other DOE Science Grid sites (Lawrence Berkeley, Argonne, Oak Ridge, and Pacific Northwest national laboratories) to test and implement security procedures and other services in the prototype Grid environment. Pre-production prototypes of various Grid services (e.g., sign-on, certification, portal, firewall, and service monitoring) have been installed and are being tested on NERSC’s computing, storage, networking, and security systems.

Working with the Particle Physics Data Grid collaboration, NERSC set up a server to test virtual organization membership schemas; feedback to Globus developers based on these tests resulted in changes in the design of the Globus Community Authorization Service. NERSC’s BRO network security monitoring system has been reconfigured to work with Globus, and the NERSC Information Management (NIM) account management system is also being made Grid-aware so that it will be able to use clients’ Public Key Infrastructure (PKI) authorization certificates.

NERSC is one of eight laboratory and university sites participating in the U.S. ATLAS Grid testbed, a prototype system to support physicists who collaborate on the ATLAS experiment at CERN’s Large Hadron Collider. The goal of the ATLAS Grid is to enable collaborators to access data and perform analysis from their home institutions. In a “data challenge” exercise of this testbed, NERSC’s PDSF successfully ran 2,500 jobs submitted over a two-week period.

Tests of distributed computation are expected to begin in the upcoming year, including a subsurface flow simulation involving computers at NERSC and Pacific Northwest National Laboratory.