Comprehensive
Scientific Support
NERSC’s comprehensive scientific
support ranges from everyday high-quality operations and client
services to direct collaboration with scientists to solve
unique computational and data management problems. Whether
the challenge involves system software, networks, algorithms,
application codes, or visualization tools, NERSC’s goal
is always the same—to help make the DOE scientific community
more productive.
Making Systems Productive
for Science
NERSC users in 2002 saw significant improvements in job turnaround
time thanks to the work of the Queue Committee, made up of
NERSC Users Group volunteers and NERSC staff. Based on the
committee’s recommendations, NERSC improved turnaround
time for interactive and debug jobs during prime time by setting
aside 5% of the Seaborg compute nodes for these jobs from
5:00 a.m. to 6:00 p.m. Pacific time, Monday through Friday.
To improve fairness, we also implemented priority aging for
regular class jobs—regular jobs in
the queue for more than 36 hours will not be preempted by
new premium jobs. In addition, we provided a new
regular_long class with a connection time of up to
24 hours for jobs using 32 nodes or less. A special mechanism
was implemented to allow science-of-scale projects to run
continuously to use their large allocations.
The NERSC networking team continues working with users at
remote sites to improve end-to-end network performance. During
the past year, the data transfer rate to and from Princeton
Plasma Physics Laboratory was increased from 700 KB/s to 2
MB/s. We achieved a 7x network performance improvement with
Fermilab, a 5x improvement with the State University of New
York at Stonybrook, and a 2.5x improvement with Brookhaven
National Laboratory. Other networking achievements included
implementation of an Access Grid node for multicast videoconferencing,
and installation of a spam filter on NERSC mail servers. An
ESnet link upgrade to OC-48 (2.4 GB/s) is expected to be in
production early in 2003.
|
|
|
|
Figure
1 The PDSF Team—Steve Chan,
Shane Canon, Cary Whitney, Iwona Sakrejda, and Tom Langley—helped
the SNO collaboration analyze data that answered longstanding
questions about solar neutrinos. (Steve now works in NERSC’s
Networking and Security Group.) |
|
NERSC’s PDSF cluster system is used by several large
high energy and nuclear physics collaborations for data analysis
and simulations, and NERSC staff work closely with these groups
to support their efforts and to improve the PDSF system (Figure
1). When data from the Sudbury Neutrino Observatory (SNO)
were analyzed on the PDSF, the results revealed that solar
neutrinos do have mass and transform from one “flavor”
to another in transit from the core of the sun to the earth.
Data from the KamLAND experiment,
also analyzed on the PDSF, confirmed these results for anti-neutrinos
generated by nuclear reactors.
One of the largest users of the PDSF and NERSC’s HPSS
archive is the Solenoidal Tracker at RHIC (STAR) collaboration,
which transfers approximately 1 TB of data per week from Brookhaven
National Laboratory to NERSC, with more than 65 TB stored
in our HPSS archive to date. Results of the STAR experiment
have been analyzed on the PDSF and published in a dozen journal
articles, with more to come. Because STAR is actively involved
in the Particle Physics Data Grid, PDSF was the first production
machine at NERSC to offer Grid services, including various
versions of Globus.
|
|
|
|
|
|
Figure
2 Harsh Anand Passi installs the
latest climate modeling codes on NERSC systems and helps
researchers use them effectively. |
|
|
|
Harsh Anand Passi, a member of NERSC’s consulting staff,
provides key support for climate modeling researchers (Figure
2). She spearheaded one of the first parallel implementations
of the netCDF library, provides initial visualization support
for climate models, and has been the technical leader for
installing several climate codes on NERSC systems, most recently
the Community Climate System Model (CCSM2). She created methods
that allow researchers to modify CCSM manageably, and developed
documentation on the code for the NERSC Web site. As a result
of her work, NERSC is the only site besides the National Center
for Atmospheric Research that runs a fully supported CCSM2.
|