PicsouGrid/Grid5000 Benchmark Results 2006-03-14
Other Results
Description
On 13 March 2007 I submitted 12 multi-node tasks with reservation to all
the available Grid5000 clusters (Grenoble was not available), each one
trying to use at least 20 nodes. The reservation was for 6:05am the
next day, Wednesday 14 March. 9 clusters at 7 sites ran the jobs successfully.
After resubmission attempts later the same day, 2 more clusters also
completed the test.
Each task was a set of simple Monte Carlo simulations, where each one takes
about 90 seconds on a "standard" desktop machine. Once the task started on the
cluster, it spread to all nodes in the OAR_NODELIST and then forked one process
per CPU.
The graphs below show a blue box for the cluster occupancy, which is
the time from the task starting on a worker node until the last node in the
nodelist has returned its results. The grey boxes signify either the task
queue time or the task data stage-out time. The red boxes show the
worker node occupancy, for each worker node. The black lines
show the life line of the core Monte Carlo algorithm.
Images
(note: these are high resolution, in order to see the life-lines for individual cores)
Discussion
This is a significant improvement to the results from 7 March. Only Lille showed 3
nodes with a clock which was one hour off.
It is interesting to see that Bordeaux has such a big performance difference
between different nodes. I need to look at the log files to see what the
actual CPUs are behind these different nodes, and to check the node load level
during the benchmark.
It is also interesting to see the delay in start time, given the reservation.
The reservation window was set for 20 minutes (although I think a bug meant
this was interpretted as 20 hours).
For Bordeaux, the first several nodes are dual AMD Opterons 248 processors (2.2
GHz). The remainder are dual Intel Xeon 3GHz processors with Hyper Threading
enabled (4 virtual processors). This means the benchmarks here aren't really
"fair". Furthermore, the nodefile supplied by OAR only listed the
number of physical processors for the Xeons. A rough "correction" would be to
divide the Bordeaux results by two, although that would also divide by two the
11 (of 51) nodes which used the AMD cores. Obviously my benchmark strategy needs
to be improved here.
Follow up
If you have any questions or would like the raw data or
images in a better format, email me at Ian.Stokes-Rees _AT_ inria.fr.
