[Bioclusters] inconsistent output from simple BLAST job

[peter_webb at agilent.com]

Hi Joe, Kums,

Per the suggestions given, and another received privately, we've set the NFS options to

rw,noac,rsize=8192,wsize=8192,timeo=14,intr

In 16 hours of testing, I've seen no errors (good!!!), but have noticed that things are running slower (bad, but preferable to the previous state of affairs).  The amount of data being moved in the tests is quite small, and processor utilization is low, so I don't have a smoking gun; I'm looking at moving files over to temporary storage on the compute nodes rather than continuously accessing them over the network.

Thanks for the help to date, if you have any comments on the options, please let me know.

Thanks,

Peter


> From: Joe Landman <landman at scalableinformatics.com>
> Subject: Re: [Bioclusters] inconsistent output from simple BLAST job
> To: bioclusters at bioinformatics.org
> Message-ID: <41A25DCD.1080500 at scalableinformatics.com>
> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
> 
> Hi Peter:
> 
>   Could you tell us what your NFS mount options are for the 
> node?  Could 
> be a case of async writes and bad flushing of output.
> 
> Joe



> 
> Message: 3
> Date: Mon, 22 Nov 2004 16:21:11 -0600 (CST)
> From: Kumaran Rajaram <kums at mpi.mpi-softtech.com>
> Subject: Re: [Bioclusters] inconsistent output from simple BLAST job
> To: "Clustering,	compute farming & distributed computing in life
> 	science informatics"	<bioclusters at bioinformatics.org>
> Message-ID: <Pine.GSO.4.58.0411221616180.27252 at mpi.mpi-softtech.com>
> Content-Type: TEXT/PLAIN; charset=US-ASCII
> 
> 
> Peter,
> 
>    Use NFSv3 and mount NFS with "noac" option. "noac" disables file
> attribute caching and should by default flush write operation to the
> disks (on the server). It checks for changes in file 
> attributes for read
> operation. The disadvantage with this option is that it 
> degrades the file
> write performance.
> 
> HTH,
> -Kums