[gmx-users] Parallel MD performance

[Mark Abraham]

On 11/06/2012 3:06 PM, delara aghaie wrote:
> Dear Gromacs users
> I have checked something today and it seems starnge to me. Doing the 
> Justin lyzozyme tutorial, I checked the performance of our parallel 
> system. Subimitting the job on 8 processors of one node, I get0.86 
> ns/daywhich seems to be very weak performance. Do you have any idea 
> about the ns/day performance of 8 processors for the same lyzozyme system?
>

When starting a new discussion, please do not reply to an existing 
email. This confuses the mailing list archive.

Parallel algorithms are not magic - they are engineered to provide 
benefits under a range of conditions. Simply observing the rate on one 
system at a number processors can give no real idea whether you are 
getting value for running in parallel. This will depend on the size of 
your simulation system and the attributes of your hardware. There is 
lots of output in the .log file about how the parallel calculation is 
running, and you will need to have read and understood a fair bit of 
chapter 3 of the manual (particularly the last sections) for that to 
make sense.

Mark
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-users/attachments/20120611/059cf577/attachment.html>