[ome-users] NFS backing store?

[Josh Moore]

Hi Jake,

On Wed, Nov 4, 2015 at 4:11 AM, Jake Carroll <jake.carroll at uq.edu.au> wrote:
> We’re currently using an XFS formatted backing store provided by a CepH pool
> for our OMERO database. It works very well!

Good to hear. Anything you can share would be appreciated. (We've all
see your youtube video!)


> We’re being asked to consider NFS as a mechanism for the transport instead.
>
> My experience early on back in OMERO 3.x days (and even early 4.x days) was
> that NFS was a very poor choice. It seemed to leave a lot of stale file
> hands active, Postgres regularly ran into issues and on the whole
> performance was not ideal.
>
> Has anything changed? Further – why was it always so problematic? Was it a
> function/semantic of the way the DB did locking workloads against its files?
> Trying to isolate why it never worked so well so that we might fix it…

On the OMERO side, not much has changed. It's possible to workaround
the NFS locking issues, but if/when we require more file locking, this
will become more difficult. That being said, if you can configure your
NFS server to obey proper posix locking semantics, then things should
be fine.

PostgreSQL is another matter. I don't think it has ever been
suggestible to run PG on NFS. If you could separate the too
(considering that the PG space requirements are much lower), you might
be able to get something working.


> Also – how are things looking from an object backend? Any further
> developments on SWIFT or S3?

Not yet, but we're always looking for motivation! :)


> Thanks all!
> -jc

Cheers,
~Josh