[Gluster-users] Tuning for Home Directories

[Joel Young]

Re home directory tuning...

Any suggestions?  We are still getting extraordinary slowness in using
gluster for home directories.  SVN checkouts to the volume are very
slow, configure scripts are very slow.

The cluster hardware is overspecced.  Each of the four nodes are
connected via 10GB ethernet, have 24-64 cores, and from 128GB to
512GB.

Thanks!

On Wed, Sep 18, 2013 at 10:44 AM, Joel Young <jdy at cryregarder.com> wrote:
> I've got a HPC type set up with four nodes and a
> distributed-replicated configuration.  Performing developer type
> actions is disfunctionally slow on the volume.  Below is the
> configuration:
>
> Volume Name: home
> Type: Distributed-Replicate
> Volume ID: 83fa39a6-6e68-4e1c-8fae-3c3e30b1bd66
> Status: Started
> Number of Bricks: 2 x 2 = 4
> Transport-type: tcp
> Bricks:
> Brick1: ir0:/lhome/gluster_home
> Brick2: ir1:/lhome/gluster_home
> Brick3: ir2:/lhome/gluster_home
> Brick4: ir3:/raid/gluster_home
> Options Reconfigured:
> server.statedump-path: /tmp
> performance.cache-size: 512MB
> performance.client-io-threads: on
>
> For example, configure runs take tens of minutes instead of fractions
> of minutes.
>
> I've tried setting lookup-unhashed to off or no without a significant
> performance change.
>
> Does anyone have any performance tunings or suggestions for making
> gluster based home directories usable for developers?
>
> Also, the documentation is unclear as to if lookup-unhashed should be
> set to yes or on, vs no or off.  Some things indicate one way, some
> the other.  boolean options should no allow multiple values to be set.
>
> Thanks!
>
> Joel