Re: [ccp4bb]: NFS Performance

[Paul Emsley <paule@chem.gla.ac.uk>]

On Tue, 2002-04-16 at 16:57, Kay Diederichs wrote:
> ***  For details on how to be removed from this list visit the  ***
> ***          CCP4 home page http://www.ccp4.ac.uk         ***
> 
> I have been experiencing exactly the same problem for a long time. This
> not only affects SOLVE, but also arp/Warp - both of which are
> long-running programs.
> 
> We work around the problem by using local (temporary) disks for these
> programs, but of course this is a major pain.

I would agree with you if you have CCP4_SCR equivalent to SCRATCH. 
However, I have found it useful if temporary results are written to
(local) CCP4_SCR but final temporary results are written to SCRATCH
(network consistent).  This does not help ccp4i users however.

> 
> I seem to remember that the problem does not exist for PGI-compiled CCP4
> binaries (the ones that used to be distributed with arp/Warp but which
> have since expired ...), which might indicate that this is a g77
> buffering problem. Can anybody confirm this?

Seems strange.

> 
> Kay
> 
> Martin Singleton wrote:
> > 
> > 
> > We have recently started using Linux (RedHat 7.2) machines in our SGI
> > based network.
> > All of our user accounts are held on an NFS exported filesystem from an
> > SGI server. This filesystem is mounted on the Linux machine as well as the
> > other SGIs. Linux sees the filesystem fine, but there appears to be a
> > major performance problem when certain CCP4 programs as well as SOLVE on
> > the Linux box requires access to this filesystem. The problem seems to
> > occur during read/writes of large binary files, which cause programs to
> > slow to a crawl, in some cases increasing the run time by two orders of
> > magnitude. 

To test disk (and nfs) performace, try bonnie

http://www.textuality.com/bonnie/

For your amusement, I attach the results for one of our GNU/Linux
clients on an unpatched SGI NFS v2 (default rsize, wsize) server: 

Paul.

bonnie.a