From: Bogdan Costescu (Bogdan.Costescu_at_IWR.Uni-Heidelberg.De)
Date: Fri Apr 22 2005 - 11:08:13 CDT

On Fri, 22 Apr 2005, Kos wrote:

> Bogdan please explain what do you mean about CHARMM problems by 'I/O
> bandwidth and passes' (examples HW)?

There are cases when several distinct parameters (f.e. some phi
angles) are required to be plotted against time. The "classical"
approach is to extract only one parameter in each pass through the
trajectory, such that for n parameters the trajectory is completely
read n times from disk/network, if the trajectory is read one frame at
a time. If the trajectory is read in memory before analysis, several
such pases can be done on the data in memory, therefore no additional
disk/network resources are used.

Of course, for systems where the available memory is larger than the
trajectory to be read, the operating system can do some caching and
provide data much faster (also from memory, but a different "pool")
than disk/network.

-- 
Bogdan Costescu
IWR - Interdisziplinaeres Zentrum fuer Wissenschaftliches Rechnen
Universitaet Heidelberg, INF 368, D-69120 Heidelberg, GERMANY
Telephone: +49 6221 54 8869, Telefax: +49 6221 54 8868
E-mail: Bogdan.Costescu_at_IWR.Uni-Heidelberg.De