[gmx-users] OpenMosix and Gromacs
leafyoung at yahoo.com
Sat Jun 30 14:11:31 CEST 2007
OpenMosix indeed provides a second model for clustering besides beuwolf
model, which uses MPI, PVM-based communication. You may run thousands of
processes and get good scaling.
I tried it with normal network, 100M Ethernet, two processes, quite bad
efficiency, 1.1-1.3x speed-up. To get OpenMosix running for high
performance, a good networking is necessary. IMHO, I don't think
OpenMosix will help in the context of gromacs or any MPI-based parallel
program because it adds another level of control over inter-process
communication which can be more directly managed by MPI library itself.
Plus, without OpenMosix, processes are fixed to be run on certain nodes
without concerns whether to migrate or not. Unless you don't have a
dedicated cluster for Gromacs, you may consider.
OpenMosix is more fit to be used for
1) building a computing farm with minimal client installation.
2) dynamic environment. The computing farm can withstand dynamic
addition and withdrawn. Nodes can be loaded to run other jobs.
3) Combing coLinux and OpenMosix and running background on Windows-based
4) huge amount of processes, each of them is running for sufficient
amount of time for OpenMosix to migrate.
5) Next-generation software to come out...
On 6/30/2007 7:17 PM, David van der Spoel wrote:
> Nicolas Sapay wrote:
>> Actually, I have used OpenMosix some time ago in a another life :)
>> All I remember is that NAMD was performing reasonably well as far as
>> you didn't ask it to write coordinates every frames. It was handy to
>> minimize/equilibrate a system or do some small computation like
>> pressure profile. As far as I remember, I was able to compute
>> something like 0.8 ns/day on 10 proc for a system of 30~20000 atoms.
> there is nothing stopping you from trying. In order for it to be
> efficient you probably need many more processes than processors, and
> IIRC that is how NAMD works. It will definitely not be more efficient
> than the gromacs development code with MPI though.
>> Sabuj Pattanayek wrote:
>>> My 0.02. Based on one of Moshe Bar's OpenMosix papers, Mosix increases
>>> the performance of MPI processes by load balancing. However I would
>>> to say that this only works for compute bound processes. If there's
>>> going to be lots of I/O between nodes Mosix won't do so well and may
>>> very well degrade performance.
>>> Did you run any benchmarks of NAMD + its parallel (CHARM?)
>>> implementation vs NAMD + CHARM + Mosix? Does it help? Last I checked
>>> Mosix userland tools and some other key functionality still don't work
>>> for 2.6 linux kernels.
>>> Nicolas Sapay wrote:
>>>> Hi everybody,
>>>> I wonder if someone has some experience with Gromacs and OpenMosix.
>>>> someone tried recently to run Gromacs in parallel with an OpenMosix
>>>> cluster? I have read some messages about that on the mailing-list but
>>>> they have been posted several years ago... Things may have changed
>>>> Additionally, I have already used sucessfully NAMD and OpenMosix (but
>>>> NAMD uses its own tool for parallel process).
>>> gmx-users mailing list gmx-users at gromacs.org
>>> Please search the archive at http://www.gromacs.org/search before
>>> Please don't post (un)subscribe requests to the list. Use the www
>>> interface or send it to gmx-users-request at gromacs.org.
>>> Can't post? Read http://www.gromacs.org/mailing_lists/users.php
More information about the gromacs.org_gmx-users