[Bio-packaging] Distributed file system and experimental cluster

[Fields, Christopher J]

Just a note on GlusterFS from our end: our IT had a test GlusterFS set up that had zfs enabled for compression, and we ran into all sorts of problems with I/O on large files.

We’re planning on performing more tests using a few common workflows on Gluster (w/o ifs) vs GPFS (our current FS, very nice but quite $$$), and possibly a few others.

chris


From: bio-packaging on behalf of Francesco Strozzi
Date: Wednesday, February 10, 2016 at 10:20 AM
To: Pjotr Prins
Cc: "bio-packaging at mailman.open-bio.org<mailto:bio-packaging at mailman.open-bio.org>"
Subject: Re: [Bio-packaging] Distributed file system and experimental cluster

Hi Pjotr,
I've used GlusterFS in the past and I think it's reasonably simple to install and configure for testing purposes. We had nice experiences with GlusterFS in mirroring mode on production systems. We tested also the striping mode although in that case we saw some bad behaviour from the server which resulted in files becoming not accessible after some time. But that was ~2 years ago so hopefully those bugs should be solved now.

Cheers
Francesco

On Wed, 10 Feb 2016 at 13:37 Pjotr Prins <pjotr.public66 at thebird.nl<mailto:pjotr.public66 at thebird.nl>> wrote:
On Wed, Feb 10, 2016 at 10:47:38AM +0100, Ricardo Wurmus wrote:
>
> Pjotr Prins <pjotr.public66 at thebird.nl<mailto:pjotr.public66 at thebird.nl>> writes:
>
> > Anyone experience with distributed file systems, such as moosefs,
> > glusterfs etc.? I want to try something reasonably easy to install and
> > manage and has striping or chunking.
>
> What do you want to achieve with a distributed file system that could
> not be achieved with centralised storage?  In the configurations that I
> worked with distributed file systems are only used for volatile scratch
> space.  Everything else (e.g. user software, home directories) uses
> either centralised storage or is node-local (e.g. the centrally managed
> system state).
>
> That said, I only ever worked on distributed file systems as a user.  I
> was never in charge of setting up a distributed file system.

With large data and many nodes the central storage quickly is the IO
bottleneck on a cluster. At least in more conventional designs.

Pj.

--
_______________________________________________
bio-packaging mailing list
bio-packaging at lists.open-bio.org<mailto:bio-packaging at lists.open-bio.org>
http://lists.open-bio.org/mailman/listinfo/bio-packaging<https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.open-2Dbio.org_mailman_listinfo_bio-2Dpackaging&d=BQMFaQ&c=8hUWFZcy2Z-Za5rBPlktOQ&r=fbHa8Njtvh9VmSnzJxiEUTW9NWDwMMwQAzhgZDO41GQ&m=GYxtZKjTsiOfMz4GCWtSC5nB6dbjD36HmxS2qbtfu1M&s=Bxb5uH2sMtxSdYBNvSiBNtXuoC9WbDmaPkLQJg7UF6A&e=>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.open-bio.org/pipermail/bio-packaging/attachments/20160210/066765be/attachment.html>