Dear Ross,
I tried to do some tests as you told me to, and the memory requirements are listed in the following:
ACE-ALA-NME gas phase 4091MB
ACE-ALA-NME igb=0 4090MB
bench.jac 104MB
bench.jac.ips 96528 96k? or 96MB? I am not sure.
bench.jac.pmemd 7985MB
Even larger memory is required for pmemd. So I attach config.h for pmemd in this mail. Maybe it can help.
Best regards,
Ye Mei
ymei.itcc.nju.edu.cn
Institute of Theoretical and Computational Chemistry
Key Laboratory of Mesoscopic Chemistry of MOE
School of Chemistry and Chemical Engineering
Nanjing University
Nanjing 210093
P.R.China
2008-04-28
======= 2008-04-28 00:52:09 Ross Walker wrote=======
>Hi Ye,
>
>What you are seeing seems very strange. I am copying this to Roberto
>Gomperts at SGI who is the expert on Amber on Altix machines. However in the
>meantime can you try running a Generalized born job instead. So just set up
>something simple like ACE ALA NME in gas phase and then run it with ntb=0
>and igb=1. This will use a different route through the code and if that uses
>a more reasonable amount of memory then we know it is related to the
>ewald/pme code. It is also possible that the code to calculate the required
>memory for such a small system is miscalculating and requesting a huge
>allocation. Can you also try a bigger system such as the JAC benchmark and
>see how this does.
>
>All the best
>Ross
>
>-----Original Message-----
>From: owner-amber.scripps.edu [mailto:owner-amber.scripps.edu] On Behalf Of
>Ye Mei
>Sent: Sunday, April 27, 2008 7:53 AM
>To: amber
>Subject: Re: RE: AMBER: memory usage
>
>Dear Ross,
>
>Thank you very much for your reply.
>I attach some files that might be helpful for solving this problem.
>As you guessed, I am using Amber9. In the attached files, please find
>config.h for the compilation. Whatever I compile amber9 statically or
>dynamically, the results are the same.
>Other info related to the compilation are:
>Intel Fortran Compiler 10.1.008
>Intel Math kernel library 10.0.1.014
>SGI native mpi
>
>What I am running is a tiny job, i.e. a ethanol in TIP3P water box. Related
>files are also attached.
>The performance is OK.
>The administrator told me that this SGI Altix 4700 uses memory as swap to
>accelerate swapping, instead of hard disk. I have no idea whether this has
>anything to do with this large virtual memory requirement. When running
>sander, little memory left for other jobs. Some CPUs cannot get enough
>memory to run jobs. So the administrator ordered me to check my amber9.
>
>Best regards,
>
>Ye Mei
>ymei.itcc.nju.edu.cn
>Institute of Theoretical and Computational Chemistry
>Key Laboratory of Mesoscopic Chemistry of MOE
>School of Chemistry and Chemical Engineering
>Nanjing University
>Nanjing 210093
>P.R.China
>2008-04-27
>
>
>======= 2008-04-27 14:10:22 Ross Walker wrote=======
>
>>Hi Ye,
>>
>>>10513 chem 25 0 15.7g 161m 57m R 100 0.2 2028:17 sander.MPI
>>
>>This doesn't look like much memory to me. 0.2% seems perfectly reasonable.
>>The 15.7GB of virtual memory seems a bit strange though. Are you sure this
>>is not some strange problem with reporting of memory usage? Why did the
>>administrator ask you to check things? Is it causing problems? The jobs are
>>all running at 100% cpu usage so it isn¨t clear that it is swapping, more
>>likely there is 15.7 GB that got swapped out immediately.
>>
>>This is AMBER 9 I assume? If so then this is weird for a 4000 atom run,
>>although I suppose if you were using Generalized Born it might use a lot,
>>but not that much. Can you post details of the actual calculation you are
>>running.
>>
>>Of course if this is an earlier version of sander which used to have static
>>memory then it is possible it was compiled with a "huge" sizes.h file that
>>requests 16GB of static memory. Then even if you run a tiny job it still
>>allocates 16GB and then of course the memory manager then swaps out the
>>15.7GB or so that is not being used. This doesn't correlate with the
>>executable being called sander.MPI though...
>>
>>Thus I am a little confused. It will probably be easier to see your input
>>script etc. And the first part of the output. Also what is the performance
>>like? Very slow or inline with what you'd expect?
>>
>>All the best
>>Ross
>>
>>
>>-----Original Message-----
>>From: owner-amber.scripps.edu [mailto:owner-amber.scripps.edu] On Behalf Of
>>Ye Mei
>>Sent: Saturday, April 26, 2008 6:58 AM
>>To: amber mailing list
>>Subject: AMBER: memory usage
>>
>>Dear amber users,
>>
>>I got quite confused about memory usage by sander.MPI running on SGI Altix.
>>As far as I know, sander does not use much memory, especially for a tiny
>>system with less than 4000 atoms. It is true on PC. But it is not the case
>>on SGI Altix 4700 running SuSE Linux Enterprise Server 10 with patch level
>>1. The top command shows that it occupies more than 15GB virtual memory as
>>following.
>> PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND
>>10513 chem 25 0 15.7g 161m 57m R 100 0.2 2028:17 sander.MPI
>>10514 chem 25 0 15.7g 120m 14m R 100 0.2 2028:06 sander.MPI
>>10515 chem 25 0 15.7g 161m 57m R 100 0.2 2028:07 sander.MPI
>>10516 chem 25 0 15.7g 160m 56m R 100 0.2 2028:19 sander.MPI
>>10517 chem 25 0 15.7g 160m 57m R 100 0.2 2028:07 sander.MPI
>>10518 chem 25 0 15.7g 158m 55m R 100 0.2 2028:19 sander.MPI
>>10519 chem 25 0 15.7g 160m 57m R 100 0.2 2028:19 sander.MPI
>>10520 chem 25 0 15.7g 159m 56m R 100 0.2 2028:07 sander.MPI
>>
>>The server administrator ordered me to check my jobs. But I have no idea
>how
>>this could happen. Does anyone know how to solve this problem?
>>BTW, jobs are managed by PBS Pro.
>>
>>
>>
>>Best regards,
>>
>>Ye Mei
>>ymei.itcc.nju.edu.cn
>>Institute of Theoretical and Computational Chemistry
>>Key Laboratory of Mesoscopic Chemistry of MOE
>>School of Chemistry and Chemical Engineering
>>Nanjing University
>>Nanjing 210093
>>P.R.China
>>2008-04-26
>>
>>-----------------------------------------------------------------------
>>The AMBER Mail Reflector
>>To post, send mail to amber.scripps.edu
>>To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>>
>>-----------------------------------------------------------------------
>>The AMBER Mail Reflector
>>To post, send mail to amber.scripps.edu
>>To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>>
>
>= = = = = = = = = = = = = = = = = = = =
>
>
>
>
>
>
>
>-----------------------------------------------------------------------
>The AMBER Mail Reflector
>To post, send mail to amber.scripps.edu
>To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>
= = = = = = = = = = = = = = = = = = = =
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
- application/octet-stream attachment: config.h
Received on Wed Apr 30 2008 - 06:07:25 PDT
