cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
Showing results for 
Search instead for 
Did you mean: 

Mech WF2.0 with Multiple Processors

Highlighted
Newbie

Mech WF2.0 with Multiple Processors

I have a new Dell 690 with 2 x Xeon 3.0 Ghz Dual Core processors. It
also has the 15K rpm SAS hard drive and 2 gig ram on Windows XP Pro. I
did a test to see what sort of performance improvement I get from having
multiple processors. I set the environment variable, MEC_NUM_THREADS,
to values of 1, 2 and 4. Here are my results:

Mec_Num_Threads Total Time (s) CPU Time (s) Improvement
1 374.02 239.81 n/a
2 348.46 247.14 6.83%
4 336.40 251.33 10.06%


I know that only part of the analysis runs in multi-threaded mode but I
would have expected a greater improvement from the extra capabilities.
Anyone have any thoughts or suggestions on how to improve performance
with multiple processors???
1 REPLY 1

Re: Mech WF2.0 with Multiple Processors

Dear Greg,

It looks to me as though your test model is too small a problem to really
stretch your machine. Only the matrix inversion (solving equations) part
of the solver is multi-threaded, so take a look at the elapsed time for
this portion of the analysis. I have been running a dual-core Opteron 64
bit workstation for about a year now and I regularly see CPU times well in
excess of the Total time, which seems odd, but the CPU time is the total
time taken by all processors.

Increasingly nowadays the limiting factor in Mechanica analyses is the I/O
time in writing results. Here's a recent example of mine:

Degrees of Freedom: 2.2 million
Total run time: 8512 sec.
Time to write results: 2120 sec (25%)

Regards,

Rod Giles
Elite Consulting Ltd.
The Mechanica Specialists

> I have a new Dell 690 with 2 x Xeon 3.0 Ghz Dual Core processors. It
> also has the 15K rpm SAS hard drive and 2 gig ram on Windows XP Pro. I
> did a test to see what sort of performance improvement I get from having
> multiple processors. I set the environment variable, MEC_NUM_THREADS,
> to values of 1, 2 and 4. Here are my results:
>
> Mec_Num_Threads Total Time (s) CPU Time (s) Improvement
> 1 374.02 239.81 n/a
> 2 348.46 247.14 6.83%
> 4 336.40 251.33 10.06%
>
>
> I know that only part of the analysis runs in multi-threaded mode but I
> would have expected a greater improvement from the extra capabilities.
> Anyone have any thoughts or suggestions on how to improve performance
> with multiple processors???
>
Announcements
LiveWorx Call For Papers Happening Now!