I benchmarked MrBayes 3.1.2 program on my cluster for two cases:

1. use local /tmp for temporary files
2. use the network shared /home/galaxy/galaxy-dist/database/tmp

MrBayes is about 10 times slower for case 2 than for case 1.  What I did was to set the network shared folder as the default but in the MrBayes wrapper, I changed the environment variable TEMP to be a local folder. 

Luobin

On Tue, Jul 26, 2011 at 10:39 AM, Peter Cock <p.j.a.cock@googlemail.com> wrote:
On Tue, Jul 26, 2011 at 5:16 PM, Duddy, John <jduddy@illumina.com> wrote:
> I can give you a very good example - if you are doing alignment and for some
> reason need to convert the input file before operating on them, such that you
> need a complete copy, /tmp may not have enough room. I have had this happen
> to me running lots of instances of an aligner, temporarily using 100G+ of temp space.
>
> I don't see the need to have a "shared" temp space, but I do see the need to
>  be able to tell the tools where you want them to put temp files.

So in your setup, the cluster nodes are not likely to have 100G+ on /tmp (i.e.
the local hard drive of the node), so you want them to use a temp folder on the
cluster shared storage?

I think needs will differ between tools - in some cases you really want
a fast local drive for temp files, and putting them on a network drive
will just kill performance. Using /tmp seems a safe default.

Is there any guidance for tool authors on where to put temp files, and
how to access any related Galaxy settings? There is nothing currently
listed here: http://wiki.g2.bx.psu.edu/Admin/Tools/Tool%20Config%20Syntax

Peter
___________________________________________________________
Please keep all replies on the list by using "reply all"
in your mail client.  To manage your subscriptions to this
and other Galaxy lists, please use the interface at:

 http://lists.bx.psu.edu/