 |
|
DAILY NEWS AND INFORMATION
FOR THE GLOBAL GRID COMMUNITY / OCTOBER 13, 2003: VOL. 2 NO. 41
|
Special Features:
NEW GRASP BENCHMARK PROBES ASSESS
GRID COMPUTING PERFORMANCE
The San Diego Supercomputer Center (SDSC), in conjunction with the Computer
Science and Engineering department at the University of California-San Diego
(UCSD), have released the Grid Assessment Probes (GRASP) software package,
which gives system architects and applications developers a simple and
easy-to-use set of probes for gaining insight into the performance and
reliability of Grid computing platforms.
Scientists who run data-intensive applications often require access to
distributed resources, data repositories and computing facilities that may be
located at various sites across the nation or around the world.
"The trend is for information from data acquisition devices -- radio
telescopes, electron microscopes, and environmental sensors -- to grow at
rates that exceed Moore's law for the increase in speed of single processors
and capacity of single disk drives," said Chaitan Baru, co-director of SDSC's
Data and Knowledge Systems (DAKS) program. "Sharing entire systems of
processors and storage resources and accessing them on demand helps to keep up
with the flood of information. Grid computing provides a promising
approach."
GRASP makes it possible to evaluate a Grid platform both quantitatively and
qualitatively. In developing GRASP, researchers at the Grid Research and
Innovation Laboratory (GRAIL) of SDSC and UCSD CSE and the Performance
Modeling and Characterization (PMaC) lab at SDSC focused on benchmarks that
were representative of data-intensive applications.
"The mission of the Grid Benchmarks Research Group is to promote the
efficient
use of Grids by defining metrics to measure the performance of Grid
applications and architectures and to rate their functionality and
efficiency," said PMaC Laboratory leader Allan Snavely. "These metrics should
facilitate good engineering practices by allowing system designers to
quantitatively compare alternative implementations. They should also give Grid
users information about systems capabilities so they can develop and tune
their applications."
As a first step towards these goals, GRAIL and PMaC have developed a set of
probes that exercise basic Grid operations by simulating simple applications.
The Grid Assessment Probes test and measure performance of basic Grid
functions including file transfers, remote execution, and Grid Information
Services response. They also can serve as simple Grid application examples and
test cases.
The SDSC and UCSD researchers have been running the probes on the GrADS
testbed, which consists of nearly 100 networked Pentium and IA-64 machines in
Tennessee, Illinois, Texas and California that run variants of the Linux
operating system. The researchers have been collecting performance data on the
testbed, including compute times, network transfer times, and Globus
middleware overhead. The results of these experiments help provide insight
into the stability, robustness, and performance of the testbed, and have led
to recommendations for future Grid software development. GRASP is outlined in
paper-and-pencil specification for several typical data-intensive application
formats.
"Running the GRASP probes provided a lot of information about the
performance
and reliability of the Grid," said Holly Dail, a programmer analyst in the
GRAIL lab. "We were able to observe bottlenecks and weak links, diagnose some
failures, and obtain performance expectations for our Grid applications."
A reference implementation and a technical report of preliminary
measurement
results from the probes are available for download at
grail.sdsc.edu/projects/grasp/. Users can run, copy, modif,
and
distribute GRASP software and its documentation for educational, research and
non-profit purposes without fee, subject to the very simple provisions listed
on the Web site.
"We invite the community to engage with us by running and reporting results
from GRASP to better characterize factors affecting the performance of Grid
applications," said Henri Casanova, director of the GRAIL lab at SDSC.
GRASP was developed under NSF "Strategic Technologies for the Internet" (STI)
grant #0230925, "Data Intensive Grid Benchmarks," with Chaitan Baru as
Principal Investigator and Allan Snavely and Henri Casanova of SDSC as Co-PIs.
This effort is part of larger-scope community activities in the Grid
Benchmarks Research group co-founded by PMaC laboratory director Snavely and
Rob Van der Wijngaart and Michael Frumkin of NASA, in colloboration with
Casanova. For more information see grail.sdsc.edu/projects/grasp/.
The GRASP research group is affiliated with the Global Grid Forum.
|