There was a typo in the params file rendering memory optimisation with 
VIADEV_WQ_SIZE ineffective. I had messed up a underscore with a hyphen as 
in VIADEV-WQ_SIZE vs VIADEV_WQ_SIZE. I just started the same job with a 
corrected parameter name.

This means that the 252 CPUs and the total-memory available are not peaked 
for performance as a grid yet. I am going to look at MemFree from 
/proc/meminfo during this run and from my previous experience on 128node 
250processor clusters, i think the largest size should be closer to 214000 
and that is where I think we will get the rmax.

Either I will do a run of that size or you could play with HPL.dat to get 
to the max efficiency.

Standing by,
-- 
;Madhu
+91 9945180001 / +91 80 41369300
--

On Wed, 22 Feb 2006, Jaime E. Combariza wrote:

> I like these results, but we may need to exchange some emails ...
>
> 1.2x10+3 Gflops for 252 cpus, correct?
>
> ----------------------------------------------------------------------------
> WR81R2R1      175000    80     9    14            2992.47          1.194e+03
> ----------------------------------------------------------------------------
> ||Ax-b||_oo / ( eps * ||A||_1  * N        ) =        0.0166907 ...... PASSED
> ||Ax-b||_oo / ( eps * ||A||_1  * ||x||_1  ) =        0.0022244 ...... PASSED
> ||Ax-b||_oo / ( eps * ||A||_oo * ||x||_oo ) =        0.0003822 ...... PASSED
> ============================================================================
>
> Finished      2 tests with the following results:
>              2 tests completed and passed residual checks,
>              0 tests completed and failed residual checks,
>              0 tests skipped because of illegal input values.
> ----------------------------------------------------------------------------
>
>
>
>
>> 1 process and another thread .... this way mvapich's own memory
>> utilisation is reduced.
>> regards,
>> --
>> ;Madhu
>> +91 9945180001 / +91 80 41369300
>> --
>>
>> On Wed, 22 Feb 2006, Jaime E. Combariza wrote:
>>
>>> Isee the otyer jobs with different p and q. I think your hostfile
>>> will use obe cpu per node but right now there are 2 processes running on
>>> quitea  few of the nodes. Is this what you expect?
>>>
>>> different cpu time as well.
>>>
>>> Jaime
>>>
>>
>
>
>