Just to add that Mike's observations in his eLife paper agree with our own
observations that after several hundred cores the efficiency of
parallelization starts going down quite a bit. Until then it's OK.
HTH,
Sjors
> Hi Bettina,
>
> I ran performance tests using relion-1.3 on Amazon's high performance
> virtual machines that were coordinated into a sun grid engine cluster
> through Amazon's elastic cloud computing network.
>
> You can find the details in Figure 4:
>
> http://elifesciences.org/content/4/e06664
>
> While there may be differences between this configuration of virtual
> machines and the Archer computing facility, this may help as a rough
> comparison guide.
>
> Hope this helps,
> Mike
>
> On Sep 9, 2015, at 11:06 AM, Bettina Boettcher
> <[log in to unmask]<mailto:[log in to unmask]>> wrote:
>
> Dear All,
>
> I am preparing an application for access to Archer (UK-naional computeing
> facility) for computing with Relion. The application requires evidence of
> the speed-up of the code on a HPC-system. I want to use Relion 1.3 or 1.4
> Did anyone measure what the actual speed-up with increasing numbers of
> cores is (e.g for relion_refine_mpi). .. I will only be able to test up to
> 56 cores on my small cluster, but you might have some numbers for more
> cores.
> If you have measured the speed-up, would you be willing to share a graph
> or table which I could add as scaling evidence, I would be most grateful.
> Maybe others face the same problem.
>
> Best wishes and many thanks
> Bettina
>
> --
> Prof. Bettina Boettcher
> University of Edinburgh
> School of Biological Sciences
> Michael Swann Building (room 334)
> Max Born Crescent
> Edinburgh EH9 3BF
> Phone: + 44 131 6505 699
> http://boettcher.bio.ed.ac.uk/<https://urldefense.proofpoint.com/v2/url?u=http-3A__boettcher.bio.ed.ac.uk_&d=BQMCAw&c=WO-RGvefibhHBZq3fL85hQ&r=P3ZikJbfwvKU1HKLXYsUylfmZWaBESuXBxGrCl7zfjo&m=BT2PJoKZaCnpNsSudLcX8wahryyp4YTiTcmq4C1ZHHA&s=y_ps1cBZGMCr7LA4UeKVG4IMJTMt-TlbtcvwSkGRzNc&e=>
>
> The University of Edinburgh is a charitable body, registered in
> Scotland, with registration number SC005336.
>
--
Sjors Scheres
MRC Laboratory of Molecular Biology
Francis Crick Avenue, Cambridge Biomedical Campus
Cambridge CB2 0QH, U.K.
tel: +44 (0)1223 267061
http://www2.mrc-lmb.cam.ac.uk/groups/scheres
|