Re: [gmx-users] Re: Question about parallazing Gromacs (Qiao Baofu)
Hi,Thanks. I have test different cpus. Our institute has two clusters: one is each node has 4 cpu (A), one is one node has only 1 cpu (B). I made different tests on the two clusters and my local computer using the same system. See the following result: A (For 1 hour) # of cpus ; MD steps 4 finished (20steps for 26:21) 8 finished (20steps for 40:57) 12 87950 20 42749 44 5962 ! B (For 1 hour) # of cpu ; MD steps 1 156991 for 56:12 2 179820 3 200,000 for 54:20 4 200,000 for 51:12c. Local(single cpu), 20 steps For 1h52:38One can see that 1. On cluster A, one nodes(4 cpu) is just as 4 times fast as my local computer.2. More than one nodes will decrease the performancs the gromacs, 3. On cluster B, the more cpu used, the faster gromacs runs. But the difference of speed is not apparent.4. Cluster B with 4 cpus is slow as half as that Cluster A with 1 node (4 cpus)I wonder if anyone can tell the bottlenack: the hardware on the cluster or gromacs? 2006/9/14, Mark Abraham [EMAIL PROTECTED]: You have to find your optimum making some tests with your settings. To do that you can start your simulation and interrupt after a while to have some data logged in the log file. Then, from the information in that log file you can estimate the time that the whole task will take and compare using more or less number of processors until you find your optimum value.Of course, that while should be at least of the order of several minutes. There is a set-up cost borne once at the start of the calculationwhich is not proportional to the length of the calculation, so you need torun long enough to get out of the time period during which it dominates the linear component.Mark___gmx-users mailing listgmx-users@gromacs.org http://www.gromacs.org/mailman/listinfo/gmx-usersPlease don't post (un)subscribe requests to the list. Use thewww interface or send it to [EMAIL PROTECTED] .Can't post? Read http://www.gromacs.org/mailing_lists/users.php-- Sincerely yours,** Baofu Qiao, PhDFrankfurt Institute for Advanced Studies** ___ gmx-users mailing listgmx-users@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 [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] Re: Question about parallazing Gromacs (Qiao Baofu)
On Thursday 14 September 2006 09:53, Mark Abraham wrote: Qiao Baofu wrote: Hi, Thanks. I have test different cpus. Our institute has two clusters: one is each node has 4 cpu (A), one is one node has only 1 cpu (B). I made different tests on the two clusters and my local computer using the same system. See the following result: A (For 1 hour) # of cpus ;MD steps 4 finished (20steps for 26:21) 8 finished (20steps for 40:57) 12 87950 20 42749 44 5962 ! B (For 1 hour) # of cpu ; MD steps 1156991 for 56:12 2179820 3200,000 for 54:20 4200,000 for 51:12 c. Local(single cpu), 20 steps For 1h52:38 One can see that 1. On cluster A, one nodes(4 cpu) is just as 4 times fast as my local computer. 2. More than one nodes will decrease the performancs the gromacs, 3. On cluster B, the more cpu used, the faster gromacs runs. But the difference of speed is not apparent. 4. Cluster B with 4 cpus is slow as half as that Cluster A with 1 node (4 cpus) I wonder if anyone can tell the bottlenack: the hardware on the cluster or gromacs? Probably your interconnects between nodes are using carrier pigeons or something :-) I expect that 1 cpu on machine A will require around four times as long as 1 4-cpu node, which you can presumably test for yourself. For next time, if you want to compare hardware like this, either use the same length of time or the same number of MD steps for all of your runs. Also when reporting runtimes, make it clear whether you are reporting walltime or some time * number_of_cpus, etc. :-) Search the mailing list i have post several times benchmark results for different systems with standard benchmark suite. http://www.gromacs.org/pipermail/gmx-developers/2006-January/001473.html Mark ___ gmx-users mailing listgmx-users@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 [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php Greetings, Florian -- --- Florian Haberl Computer-Chemie-Centrum Universitaet Erlangen/ Nuernberg Naegelsbachstr 25 D-91052 Erlangen Telephone: +49(0) − 9131 − 85 26581 Mailto: florian.haberl AT chemie.uni-erlangen.de --- ___ gmx-users mailing listgmx-users@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 [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] Re: Question about parallazing Gromacs (Qiao Baofu)
2006/9/14, Mark Abraham [EMAIL PROTECTED]: Probably your interconnects between nodes are using carrier pigeons orsomething :-) I expect that 1 cpu on machine A will require around fourtimes as long as 1 4-cpu node, which you can presumably test for yourself. It is forbidden to run only one cpu on the cluster A in my inisititute. For next time, if you want to compare hardware like this, either use thesame length of time or the same number of MD steps for all of your runs.Also when reporting runtimes, make it clear whether you are reporting walltime or some time * number_of_cpus, etc. :-)For all the jobs (except the one on my local computer) , I set walltime=1hour, and nsteps= 200,000,dt=0.001.The running time are taken from the end of the .log file. See the following example. NODE (s) Real (s) (%) Time: 1581.000 1581.000 100.0 26:21 (Mnbf/s) (GFlops) (ns/day) (hour/ns)Performance: 56.376 4.515 10.930 2.196Mark___ gmx-users mailing listgmx-users@gromacs.orghttp://www.gromacs.org/mailman/listinfo/gmx-usersPlease don't post (un)subscribe requests to the list. Use the www interface or send it to [EMAIL PROTECTED].Can't post? Read http://www.gromacs.org/mailing_lists/users.php -- Sincerely yours,**Baofu Qiao, PhDFrankfurt Institute for Advanced StudiesMax-von-Laue-Str. 160438 Frankfurt am Main, Germany TEL:+49-69-7984-7529 ** ___ gmx-users mailing listgmx-users@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 [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
[gmx-users] Re: Question about parallazing Gromacs (Qiao Baofu)
Well, there is a common misconception with parallel computing. Usually, you will have an optimun number of processors that guarantees the best performance. More or less than that number will result in a decreased performance and longer computation times. The optimum number of processors will depend on the particular problem and the hardware/software configuration of your cluster, but for instance, in my case for docking experiments I've found that 4 cpu's is Ok. If I try to use more than 4 cpus the performance is worst. The same is for less than 4 cpu's. You have to find your optimum making some tests with your settings. To do that you can start your simulation and interrupt after a while to have some data logged in the log file. Then, from the information in that log file you can estimate the time that the whole task will take and compare using more or less number of processors until you find your optimum value. I hope it helps. Regards, César.- -- Message: 1 Date: Wed, 13 Sep 2006 12:42:37 +0200 From: Qiao Baofu [EMAIL PROTECTED] Subject: [gmx-users] Question about parallazing Gromacs To: gmx-users@gromacs.org Message-ID: [EMAIL PROTECTED] Content-Type: text/plain; charset=iso-8859-1 Hi all, I have a question about parallazing gromacs: I run the same system on a cluster of my institute and my local computer, Cluster:* *dual processor boards AMD Opteron 270 (Dual-Core), 2.0 GHz Local computer: AMD X86-64 Cpu, double precision 1. The cluster (nodes=3:ppn=4) runs 87950 MD steps for one hour 2. The cluster (nodes=5:ppn=4) runs 42749 MD steps for one hour 3. The cluster (nodes=11:ppn=4) runs 5962 MD steps for one hour 3. My local computer runs 179090 MD steps For 1hour 51 mintues. It is verry strange that the more cpus I use, the slowest the gromacs runs.!! Who knows what's wrong with my job? And for paralleled gromacs, how many cpus is prefered? The grompp command is: grompp -np 12 -o md3.mdp -c md3in.gro -p MCl.top -o md3.tpr The following is one of the the job scripts on the cluster: # # MD NTP(BerendsenBerendsen, T=425P=1bar),200ps tau_p=4 # # #!/bin/bash #PBS -N md3 # #PBS -l walltime=01:00:00,nodes=3:ppn=4 # #PBS -m abe # #PBS -o md3.out # #PBS -e md3.err # # cd /work/fias/qiao/time_checking/nodes3/ /usr/local/Cluster-Apps/lam/gcc/64/7.1.1/bin/lamboot $PBS_NODEFILE /usr/local/Cluster-Apps/lam/gcc/64/7.1.1/bin/mpirun -np 12 mdrun -v -s md3.tpr -x md3 -e md3 -c md3 -g md3 exit 0 -- Sincerely yours, ** Baofu Qiao, PhD Frankfurt Institute for Advanced Studies ** -- next part -- An HTML attachment was scrubbed... URL: http://www.gromacs.org/pipermail/gmx-users/attachments/20060913/fdee271a/attachment-0001.html ___ gmx-users mailing listgmx-users@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 [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php
Re: [gmx-users] Re: Question about parallazing Gromacs (Qiao Baofu)
You have to find your optimum making some tests with your settings. To do that you can start your simulation and interrupt after a while to have some data logged in the log file. Then, from the information in that log file you can estimate the time that the whole task will take and compare using more or less number of processors until you find your optimum value. Of course, that while should be at least of the order of several minutes. There is a set-up cost borne once at the start of the calculation which is not proportional to the length of the calculation, so you need to run long enough to get out of the time period during which it dominates the linear component. Mark ___ gmx-users mailing listgmx-users@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 [EMAIL PROTECTED] Can't post? Read http://www.gromacs.org/mailing_lists/users.php