DaimlerChrysler loses less Jobs on their Linux and Itanium Clusters
John Picklo, Manager, Mainframes and High Performance Computing
at DaimlerChrysler in Detroit stated in his talk at the
Clusterworld Conference 2003 (San Jose, June 2003) that
before they introduced Linux-Pentium- and HP-UX-Itanium-clusters they had a job
loss rate of 0,1% due to system failure.
Introducing Linux clusters with a total of 472 CPUs
and 192 Itanium-CPUs they where able to reduce the
job loss rate due to system failure to a half of that
What are these clusters doing and how many
CPUs do they have in HPC?
In June 2003 DaimlerChrysler US had the following
clusters running (for full size please click on the
in addition to their RISC-systems :
172 Node Pentium Cluster (344 CPUs) Xeon 2.2Ghz, 2.8Ghz from
IBM running Linux
96 Node Itanium Cluster (192 CPUs) from HP running HP-UX
2 x 32 Node Pentium Cluster (total: 128 CPUs) Xeon 2.8Ghz
from IBM running Linux
SGI: 12 Origin 300 (384 CPUs)
4 Origin 3000 (256 CPUs)
2 Origin 2000 (128 CPUs)
- Impact- and NVH-simulations:
6 Superdomes (352 CPUs) from HP
For Impact-/Crash-Simulations as well as NVH-calculations
DaimlerChrysler US has added
Linux and Itanium clusters with a total of 536 CPUs. Compared to
the 352 RISC-CPUs used in that area there are about 50% more CPUs
available in clusters for this application area.
CFD-calculations at DaimlerChrysler US are
mainly run on SGI-RISC-systems with a total of 786 CPUs.
The overall proportion of RISC to cluster CPUs
is at 1120 to 664 (169%).
What about price and performance of clusters compared to SMP ?
- Pentium Linux Cluster:
20% performance improvement over equivalent number of RISC processors
40% cost benefit
Itanium HP-UX Cluster:
50% performance improvement over RISC
Lessons learned ?
Can you fly below the radar?