Re: Scaling of Sander on Linux Clusters (fwd)

From: Pratul Agarwal <pratulka_at_rocketmail.com>
Date: Mon 26 Aug 2002 17:32:24 -0700 (PDT)

Yes I have seen this behavior before (not with sander though). As the
CPUs are becoming faster the network communication (latency and actual
transport time) is starting to become the bottleneck.

Try a simple check, run the jobs on multiple CPUs and using a system
monitoring command (e.g. uptime or top) see how much of CPUs are being
used. If the usage is significantly less than 100% then your system
might be hitting the curve where transport over the network has become
the bottleneck.

For the same cluster setup a bigger system will show better performance
(assuming that other things in the setup are okay and the network is
the bottleneck).

Hope this helps.

--- amber_at_heimdal.compchem.ucsf.edu wrote:
>
>
> ---------- Forwarded message ----------
> Date: Mon, 26 Aug 2002 14:09:25 +0200
> From: Thomas Steinbrecher
> <thomas.steinbrecher_at_physchem.uni-freiburg.de>
> To: amber_at_heimdal.compchem.ucsf.edu
> Subject: Scaling of Sander on Linux Clusters
>
> Dear AMBER users,
>
> I have a question about the tutorials provided on the AMBER
> homepage and about the performance of parallel sander in
> general.
>
> While running the first tutorial, the DNA 10mer in a box of
> water, on our Linux cluster (Dual Athlon 1600MHz CPU's with
> a 100 Mbit network, not used by other users), I saw that
> the calculation takes longer if I use more CPUs.
>
> Both the minimization and the molecular dynamics run of the
> water molecules with restraints on the DNA are completed
> faster on two nodes (4CPUs) tha on 4 nodes (8CPUs).
>
> (Minimization takes 370 sec on 4 CPUs and 530 sec on 8
> CPUs, dynamics takes 6000 and 7400 sec respectively. If I
> use only one node with 2 CPUs, they times are 413 and 6500
> sec respectively)
>
> Has anyone experienced this behaviour and if yes, is it to
> be expected? I expected about half the calculation times
> with double number of CPUs.
>
> If there is an optimum number of CPUs to assign to a given
> problem, is there a way to guess it from the imput data
> (from number of atoms/molecules of the system)?
>
> If anybody has tried the tutorials recently, could you
> share your experience how long the different calculations
> took on different systems?
>
> Regards,
>
> Thomas
>


__________________________________________________
Do You Yahoo!?
Yahoo! Finance - Get real-time stock quotes
http://finance.yahoo.com
Received on Mon Aug 26 2002 - 17:32:24 PDT
Custom Search