Gluster users - Nov 2012 - Performance after failover

Hey, all.

I'm currently trying out GlusterFS 3.3.
I've got two servers and four clients, all on separate boxes.
I've got a Distributed-Replicated volume with 4 bricks, two from each
server,
and I'm using the FUSE client.

I was trying out failover, currently testing for reads.
I was reading a big file, using iftop to see which server was actually
being read from.
I put up an iptables firewall on the active server, and the client hung, as
expected.
After some time, I guess about 30 seconds, the failover kicked in and the
operations
went to the other server, resuming the read.

However, after going to the other server, throughput went from ~100MB/s to
~25MB/s.
Kicking off a new read after the previous one had succeeded went straight
to the second server,
with throughput still at ~25MB/s.

Is this the expected behavior for clients that have failed over from one
server to another ?

torbjorn at srv18:~$ sudo gluster volume$
Volume Name: testvol$
Type: Distributed-Replicate$
Volume ID: 90636b5d-0d57-483c-bbfd-c0cdab2adaaa$
Status: Started$
Number of Bricks: 2 x 2 = 4$
Transport-type: tcp$
Bricks:$
Brick1: srv18.trollweb.net:/srv/gluster/brick0$
Brick2: srv17.trollweb.net:/srv/gluster/brick0$
Brick3: srv18.trollweb.net:/srv/gluster/brick1$
Brick4: srv17.trollweb.net:/srv/gluster/brick1$


--
Vennlig hilsen
Torbj?rn Thorsen

Trollweb Solutions AS
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20121127/7c780a2e/attachment.html>

This is a self-reply.

I've not been able to reproduce the situation, and it just might be my
very own fault.
Quite possible that misunderstanding and monitoring issues were the
root causes of my problem.

On Tue, Nov 27, 2012 at 4:44 PM, Torbj?rn Thorsen <torbjorn at
trollweb.no> wrote:
> Hey, all.
>
> I'm currently trying out GlusterFS 3.3.
> I've got two servers and four clients, all on separate boxes.
> I've got a Distributed-Replicated volume with 4 bricks, two from each
> server,
> and I'm using the FUSE client.
>
> I was trying out failover, currently testing for reads.
> I was reading a big file, using iftop to see which server was actually
being
> read from.
> I put up an iptables firewall on the active server, and the client hung, as
> expected.
> After some time, I guess about 30 seconds, the failover kicked in and the
> operations
> went to the other server, resuming the read.
>
> However, after going to the other server, throughput went from ~100MB/s to
> ~25MB/s.
> Kicking off a new read after the previous one had succeeded went straight
to
> the second server,
> with throughput still at ~25MB/s.
>
> Is this the expected behavior for clients that have failed over from one
> server to another ?
>
> torbjorn at srv18:~$ sudo gluster volume$
> Volume Name: testvol$
> Type: Distributed-Replicate$
> Volume ID: 90636b5d-0d57-483c-bbfd-c0cdab2adaaa$
> Status: Started$
> Number of Bricks: 2 x 2 = 4$
> Transport-type: tcp$
> Bricks:$
> Brick1: srv18.trollweb.net:/srv/gluster/brick0$
> Brick2: srv17.trollweb.net:/srv/gluster/brick0$
> Brick3: srv18.trollweb.net:/srv/gluster/brick1$
> Brick4: srv17.trollweb.net:/srv/gluster/brick1$
>
>
> --
> Vennlig hilsen
> Torbj?rn Thorsen
>
> Trollweb Solutions AS


-- 
Vennlig hilsen
Torbj?rn Thorsen
Utvikler / driftstekniker

Trollweb Solutions AS
- Professional Magento Partner
www.trollweb.no

Telefon dagtid: +47 51215300
Telefon kveld/helg: For kunder med Serviceavtale

Bes?ksadresse: Luramyrveien 40, 4313 Sandnes
Postadresse: Maurholen 57, 4316 Sandnes

Husk at alle v?re standard-vilk?r alltid er gjeldende