ADCIRC has been optimized for both vector and parallel computer architectures.
The parallel capabilities of ADCIRC are based on a domain decomposition
strategy with information at sub-domain interfaces being passed using
the MPI library. The resulting speedup is linear or super linear for
large problems. For example, for a 300,000 node grid with 2 second time
steps, the wall clock time is about 2 hours per 24-hr
day of simulation using 128 processors on a CRAY T3E. On a 1.8 GHz Xeon processor based
machine, performance is about 4 times faster with the same number of
processors. Performance will be linearly proportional to the number of
nodes, time steps, and processors. On your single processor PC such a
run might take about 10 days computer time!
The SMS ADCIRC package contains all the necessary tools and
interfaces to run the ADCIRC model on a PC. To be able to use the SMS
package with
data from a MPI version of the ADCIRC model requires you to purchase
the ADCIRC model license that is furnished with the SMS system. SMS does
not run on an MPI platform (e.g., a LINUX cluster), but pre- and post-processing
activities can be carried out on your PC giving you the gridding and
visualization power you need to develop and analyze information to and
from a “batch” run on the MPI.
Veri-Tech, Inc. is now offering MPI (multi-cpu) ADCIRC licenses and
services to tailor the ADCIRC system to your MPI platform. Call [601-636-1454],
FAX [601-638-5641] or to
find out more information.
