| |
by Ulrich Viersbach, Ford-Werke Aktiengesellschaft;
Zafer Tastan and Mike Hesse, VISTEON Cologne, Germany
The automotive industry is under pressure to continually improve quality
while reducing cost. To enhance operating efficiencies, Ford Europe and
VISTEON, an enterprise of Ford Motor Company, jointly sought to utilize
idle cycles on their large numbers of UNIX workstations, with an initial
focus on parallel CFD calculations.
Parallel Computing on Networked Workstations
VISTEON uses FLUENT/UNS to simulate automotive heating, ventilation, and
air conditioning units. These simulations are characterized by complex airflow
(see figure below) and heat transfer phenomena.
With typical model sizes growing beyond 500,000 elements, CFD solutions
require substantial computing resources. Powerful compute servers were
previously required to achieve acceptable turnaround times. Now, improved
hardware and workload management software, and more advanced distributed-memory
parallel CAE codes have made it possible to move some calculations onto
workstation clusters.1 VISTEON Germany
has proven this by running most of its FLUENT/UNS calculations using the
distributed-memory parallel version of the code on a cluster of Hewlett-Packard
C180 workstations. To further enhance productivity, VISTEON integrated
FLUENT/UNS with the workload management capabilities of Platform Computing's
LSF software.
LSF Load Management
VISTEON engineers and Ford Vehicle CAE Systems worked jointly with Science
& Computing, the German distributor of LSF, to write scripts that seamlessly
integrate LSF with the FLUENT/UNS application environment. As a result,
FLUENT/UNS jobs and other compute-intensive applications share the clustered
computing resources reliably and transparently. For each analysis, the optimum
number of workstations in a parallel run is automatically selected based
on the size of the CAE model
Additionally, all jobs that do not finish overnight are automatically
checkpointed in the morning. This frees up critical computing resources
during business hours. Most of the workstations in the cluster are heavily
used interactively by the CAD and CAE department during the day. In the
evening, the checkpointed jobs are resumed automatically. The capability
to recover from a workstation crash will also be supported. LSF's checkpointing
and migration capabilities will be used to ensure that running jobs are
resumed elsewhere as quickly as possible.
Combining FLUENT/UNS and Platform's LSF Suite has resulted in multiple
business benefits, including higher efficiency and productivity, as well
as reduced CAE calculation costs for VISTEON.
1 U.Viersbach, M. Hesse, G. Zellemann, J. Schlanke. "Parallelization:
Readiness of Workstation Clustering for Computer Aided Engineering" Presentation
at HIPer'97 (high Performance Computing on Hewlett-Packard Systems), November
1997.
View of vehicle air-handling unit, maximum air conditioning
selected; x-direction velocity contours shown in cross section
|
|
|
