[Bioclusters] Need NFS with local disk based caching, CODA?

Mark Rae bioclusters@bioinformatics.org
Fri, 24 Oct 2003 09:36:12 +0100

On Fri, Oct 24, 2003 at 05:43:18PM +1000, Michael.James@csiro.au wrote:
> Our bioinformatic database files are outgrowing
>  the nodes capacity to store them,  (~24Gig).
> ...
> We need something that provides a network file system
>  but with local caching.   24 Gig will be heaps of space
>  to keep a good selection of the popular databases locally.
> ...
> And if it's no good, what are the alternatives?
> What are other clusters using?


We have recently installed a cluster on IBM HS-20 blades
and we are using GPFS for our filesystems. 

There are two configurations which we are trying.

The first is with each chassis of 14 blades having 
Fibre storage using a 2.4TB Nexsan unit. This
is partitioned to give 4x600GB filesystems.

The second is using GPFS to pool the on-board IDE
disks within each chassis to produge a single shared 
600GB filesystem.

In both cases the GPFS network traffic is directed over 
a private gigabit link. (Each blade has 2x1Gb network ports)
These filesystems are used for storing static datasets which
consist mainly of blast databases. In this 'read-only'
configuration GPFS performs as well as local disk,
however it doesn't cope as well with mutiple machines
writing to the same filesystem at the same time.

We have also tried the Sistina GFS, and found it to
perform better for small numbers of machines doing
both reads and writes.


Mark Rae
Informatics Systems Group
The Wellcome Trust Sanger Institute
Tel: +44 (0)1223 834244 x4880    Fax: +44 (0)1223 494919