|
Jonas Larsson, Volvo Aero Corporation, Trollhättan, Sweden
“We are extremely satisfied with FLUENT’s stability
and performance on our new 150 CPU Linux cluster. Over the three years
Volvo Aero has been using Linux clusters, Fluent has consistently met
and exceeded all our expectations. By switching to running FLUENT on
Linux clusters, we have been able to increase our computational resources
by a factor of 10.”
– Peter Emvin, Ph.D., Manager,
Aero and Thermodynamics, Volvo Aero Corporation
Jonas Larsson in front of the 150 CPU Linux cluster
By switching to Linux clusters, the CFD group at Volvo Aero Corporation
has been able to increase their computational resources by a factor of
ten with a reduced hardware budget. The transition from expensive parallel
UNIX machines to large Linux clusters has been a tremendous success, and
has led to huge improvements both in quality and leadtime for all CFD
work done.
The CFD group at Volvo Aero were pioneers in using Linux clusters. They
bought their first Linux cluster three years ago, and today have more
than 150 CPUs in the cluster, which is used only for CFD simulations using
FLUENT and their inhouse CFD code, VolSol. The CFD engineers are very
happy with the new computing environment. Stability and performance with
FLUENT and VolSol have been markedly better than on their old UNIX servers.
Because the engineers were already familiar with the UNIX environment,
the migration to Linux has gone smoothly. UNIX desktop machines are still
used for most pre and post-processing work.
A multi-stage axial compressor simulation 
An air-intake simulation of a Swedish fighter jet
Volvo Aero Corporation designs and manufactures components for military
jet engines, commercial jet engines, and rocket engines. CFD plays an
important role in all of these areas and has traditionally been a very
strong discipline at Volvo Aero. Most of the work is performed at the
CFD Center of Excellence, a leading engineering department that has a
long history of CFD experience, and which serves all business units of
Volvo Aero. Today there are twenty-four engineers; one adjunct professor,
twelve PhDs, and eleven MScs. The cluster is used only by this group and
has made it possible for them to run a whole new class of problems.
Transient, multistage turbomachinery simulations with several million
cells are now easily and routinely run using parallel processing on the
cluster.
When the cluster was first assembled the philosophy was to use as many
standard, off-the-shelf components as possible. The compute nodes are
normal desktop PCs and the network is normal 100Mbs, switched Ethernet.
A faster network or non-standard nodes can easily double the costs. Using
standard components also makes it much easier to maintain and upgrade
the cluster, since most companies already have a well-established channel
for buying and maintaining their desktop PCs. New nodes can easily be
added as the need arises and old slow nodes can be removed and reused
as desktop office PCs.
The switch to Linux clusters has also eliminated the need for a queue
system. The only type of scheduling used now is a script that displays
the cluster load on a web page. This allows users to select available
CPUs on an as-needed basis. With today’s low cost per CPU, it makes
more sense to buy new nodes as the need arises, rather than force users
to wait for CPU in a queue system.
With more than three years of experience running CFD on large Linux
clusters, Volvo Aero Corporation has no doubt that this is the computing
platform of the future. Volvo Aero has also started to replace their desktop
UNIX machines with Linux machines – creating a homogenous, low-cost/high-performance
computing environment that can scale to any future needs.
|
