> -----Original Message-----
> From: GRIDPP2: Deployment and support of SRM and local storage management
>
> > What's a sensible ballpark figure for that? And how gracefully does it
> > fail in the event of a full filesystem?
>
> At the Tier1, We run fsprobe locally on each disk server, using a 2 GB
> file in each of the CASTOR data partitions (three in total).
>
> >> In principle, I'd be interested at QMUL - though with lustre we'd
> >> need to make special arrangements I think.
>
> In theory, you can specify any valid path to fsprobe so if you can mount
> it, you can run fsprobe on it. I've never tried it on remote file systems
> so it would be interesting to try.
>
So presumably a single fsprobe install can be configured to test
multiple filesystems (in your case three CASTOR backing one), so
if a lustre client could be set up to have a directory structure
something like:
/fsprobe/ost1
/fsprobe/ost2
/fsprobe/ost3
etc.
with each directory being routed within lustre to a specific storage
target, then that single fsprobe install would be able to test
everything.
Also how much IO load does it generate (on a rough scale of 'hardly any'
to 'quite a lot, actually'?
Ewan
|