[gmx-users] RE: Gromacs benchmark --poor performance on XEONmachine

Erik Lindahl lindahl at csb.stanford.edu
Tue Oct 21 05:56:02 CEST 2003


You probably want to disable hyperthreading in the bios, or use the  
"noht" option to Linux.

  Hyperthreading can really screw up performance for multiple jobs, and  
in the very few cases where it helps (web & fileservers, not  
computation) it only gets you 5-10% extra. It might improve in the  
future, but in my opinion it is just way too much trouble without much  
advantage.

There are two problems with HT:

1. If you have a dual machine with 4 virtual CPUs and start 2 jobs,  
they might both be allocated on the first physical CPU while the second  
one is idle. This was a problem at least up to redhat 8.0, but it is  
rumored that Linux 2.6 improves it.

2. Even if you start 4 jobs, you don't have 4 real CPUs. The 4 jobs  
will be swapped in and out on the two virtual CPUs. Unless they get  
exactly the same amount of time, you will get problems with bad load  
balancing.

In summary: HT might work better when we roll out multithreading, but  
"just say no" for independent or MPI jobs :-)

Cheers,

Erik



On Monday, October 20, 2003, at 08:33 PM, Shejith P.P wrote:

> I had installed gromacs on a dual XEON( 2.4GHz ) machine +RedHat 9.0  
> and I was using np=4( I had enabled hyperthreading).  But I was  
> getting Gfloaps 1.5 and Mnb/f when I was running D.DPPC  -benchmark.  
> This is very low value when we are comparing with the results posted  
> in web site..
>    I had installed fftw -foating point and enabled mpi and I am using  
> LAM/MPI 7.0 .
> I am not able to find out where I am making mistake. Can anyone tell  
> me any hints.
>
> Rgds
> SPP
>
>
> -----Original Message-----
> From: gmx-users-admin at gromacs.org  
> [mailto:gmx-users-admin at gromacs.org]On
> Behalf Of David van der Spoel
> Sent: Monday, October 20, 2003 5:09 PM
> To: gmx-users at gromacs.org
> Subject: Re: [gmx-users] RE: Gromacs benchmark --poor performance on
> XEONmachine
>
>
> On Mon, 2003-10-20 at 13:28, Shejith P.P wrote:
>> Hi,
>>  Can anyone  give me the detailed installtion procedure for gromacs  
>> for parallel environment.
>>
> it's the same as te normal installation (see website)
>
>> Regards
>> Shejith
>>
>>
>>  -----Original Message-----
>> From: 	Shejith P.P [mailto:shejithp at hclinsys.com]
>> Sent:	Monday, October 20, 2003 11:01 AM
>> To:	'gmx-users at gromacs.org'
>> Subject:	Gromacs benchmark --poor performance on XEON machine
>>
>> Hi,
>>     I am running gromacs on P4 Xeon machine 2.4GHz. ( 1GB memeory ) -  
>> I am using a dual CPU machine.
>>
>> I have installed fftw(2.1.3) and gromacs (3.1.4 ) on my machine and I  
>> am running benchmarking using D.DPPC
>>    I have enabled hyperthreading and I was running 4 LAM on my  
>> machine.
>>  And I hav erun gromacs for 4 CPUS and executed the mdrun command for  
>> 4 nodes.
>>
>> But I am getting a very low value compared to the one I have seen in  
>> gromacs site.   I got Mnbf/s of 45 where in website It was 286 for  
>> dula xeon 2.2GHz.
>>    I am afraid whether I am making any mistakes in installton  
>> itself....can enyone send me the detailed installtion procedure for  
>> gromacs  for  parallel environment.
>>
>>
>> TIA
>> Regards
>> Shejth
>>
>> ÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿÿà˝î±êì˙¨¥…x%…Ë`˝î±êìfl
>> è˙§,fl…à∫€iÿÿ›Ãà®≈˚rÏè®æj)fjå…Ëbˇú?≠lÿºÇ«°ù^jǡ¢휗- 
>> fléÿ”æìr¸˝z·ª’ë-”Ú-∑éb”ßÔ±ëa{0…{^–ö˛z…ìzwb¶Ú ˝î±êìfl·ª’ë-fl
>> è˙§,fl…à
> -- 
> David.
> _______________________________________________________________________ 
> _
> David van der Spoel, PhD, Assist. Prof., Molecular Biophysics group,
> Dept. of Cell and Molecular Biology, Uppsala University.
> Husargatan 3, Box 596,  	75124 Uppsala, Sweden
> phone:	46 18 471 4205		fax: 46 18 511 755
> spoel at xray.bmc.uu.se	spoel at gromacs.org   http://xray.bmc.uu.se/~spoel
> +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ 
> +
>
> _______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to  
> gmx-users- 
> request at gromacs.org._______________________________________________
> gmx-users mailing list
> gmx-users at gromacs.org
> http://www.gromacs.org/mailman/listinfo/gmx-users
> Please don't post (un)subscribe requests to the list. Use the
> www interface or send it to gmx-users-request at gromacs.org.




More information about the gromacs.org_gmx-users mailing list