Gluster users - Aug 2011 - Gluster server and multi-homing, again

Hello.  I want to achieve the following with my Gluster configuration.

Each server has two fast local RAID partitions (let's call them /data1
and /data2) and two 10GigE network ports (let's call them eth1 and
eth2).

I want all communication with /data1 to happen via eth1 and all
communication with /data2 to happen via eth2.

My application has just a handful of clients and servers, but I need
the highest possible performance for reading and writing single large
(100-1000 GiB) files.  This is why I want to segregate the traffic
between the two 10GigE interfaces.

All of my clients will be mounting the cluster using the gluster native client.

I think I can do this simply by running two glusterd processes on each
server, having them use separate /etc/glusterd working directories,
each binding to a different interface and serving up a different
partition.  Except I do not know how to tell glusterd to read from
(e.g.) /etc/glusterd1 instead of /etc/glusterd.

Could someone tell me how to arrange this, short of hacking the source
code to build two different glusterd binaries?  Or suggest another way
to achieve my objective?

Thanks.

 - Pat

On Thu, Aug 11, 2011 at 11:48 AM, Patrick J. LoPresti
<lopresti at gmail.com> wrote:
> Hello. ?I want to achieve the following with my Gluster configuration.
>
> Each server has two fast local RAID partitions (let's call them /data1
> and /data2) and two 10GigE network ports (let's call them eth1 and
> eth2).
>
> I want all communication with /data1 to happen via eth1 and all
> communication with /data2 to happen via eth2.
Assuming these bricks can be in different volumes.

I haven't done it yet, somone else might have more experience, but I
think if you created 2 storage trusted pools on separate IPs it should
work.
>
> My application has just a handful of clients and servers, but I need
> the highest possible performance for reading and writing single large
> (100-1000 GiB) files. ?This is why I want to segregate the traffic
> between the two 10GigE interfaces.
>
> All of my clients will be mounting the cluster using the gluster native
client.
>
> I think I can do this simply by running two glusterd processes on each
> server, having them use separate /etc/glusterd working directories,
> each binding to a different interface and serving up a different
> partition. ?Except I do not know how to tell glusterd to read from
> (e.g.) /etc/glusterd1 instead of /etc/glusterd.
>
> Could someone tell me how to arrange this, short of hacking the source
> code to build two different glusterd binaries? ?Or suggest another way
> to achieve my objective?
>
> Thanks.
>
> ?- Pat
> _______________________________________________
> Gluster-users mailing list
> Gluster-users at gluster.org
> http://gluster.org/cgi-bin/mailman/listinfo/gluster-users
>