Prototype
of a NOAA Computational Grid
Dan
Schaffer
Forecast Systems Lab, OAR
NOAA
scientific research is supported by thousands of computers scattered
throughout the laboratories ranging from individual workstations
to large-scale clusters. Most of these resources are only available
to local users. A lack of connectivity inhibits sharing between
or even within labs. As a result, huge numbers of compute cycles
go unused and opportunities are lost to increase the scale of
scientific modeling. Meanwhile, advances in network backbone technologies
and middleware offer a new opportunity to share these resources.
Organizations such as NASA and NSF provide seamless, secure shared
access to member computers via a computational grid constructed
from these technologies. A computational grid is analogous to
an electrical power grid in the sense that, once plugged in, a
user theoretically has access to resources provided anywhere on
the grid.
The
Forecast Systems Laboratory (FSL) is leading an effort to setup
a prototype NOAA computational grid using Globus middleware. From
various funding sources including an FY ’03 NOAA HPCC grant,
we have developed a rudimentary grid consisting of one Intel Linux
machine located at the Pacific Marine Environmental Laboratory
(PMEL) and several others situated at FSL. A version of the Weather
Research and Forecast model has been coupled with the Regional
Ocean Modeling System (ROMS) over this grid. One model runs at
PMEL, the other at FSL and the boundary conditions are exchanged
over the Abilene network connection between the two labs. Experiments
show that the communication bandwidth and latency are such that
it is quite feasible to execute this loosely coupled model over
the grid. FSL has submitted a proposal for NOAA HPCC FY ’04
funding to expand the grid into a full prototype including clusters
and individual nodes from PMEL, FSL and the Geophysical Fluid
Dynamics Laboratory (GFDL).