Gluster users - Jun 2012 - glusterfs volume as a massively shared storage for VM images

Hi all,

I am looking for a solution, like, I hope Glusterfs can fit into it, that
is,
something that allows me to do live migrations of virtual machines from one
compute node to another (KVM, OpenStack).

And so I found an article about the Glusterfs-OpenStack-Connector, which
was just a basic Glusterfs setup that shared the /var/lib/nova/instances
directory across all compute nodes. This is necessary to allow virtual
machines to migrate within milliseconds from one compute node to another,
as all nodes already share every data.

Now my question is, how well does Glusterfs scale when you're having about
50+ compute nodes (on which you're about to run virtual machines). What
kind of setup would you recommend, to actually not suffer in runtime
performance, networking iops, nor in availability.

Many thanks in advance,
Christian Parpart.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20120611/e912b46f/attachment.html>

Hi Christian,

In theory it should work, but ability to properly run VMs on Gluster is
something relatively new due the improvements on granular healing so I don't
think it has been extended tested.
I wasn't able to find any people using it in production and those I heard
are using for testing. I tried myself here to use it with VMware and could never
get it working, some problem with NFS on Gluster side.

With regards performance not sure how long you have been on this mail list, but
have look for the last emails Brian Candler sent with his results for KVM VMs,
they don't seem very promising as it stands.

Fernando

From: gluster-users-bounces at gluster.org [mailto:gluster-users-bounces at
gluster.org] On Behalf Of Christian Parpart
Sent: 11 June 2012 01:11
To: gluster-users at gluster.org
Subject: [Gluster-users] glusterfs volume as a massively shared storage for VM
images

Hi all,

I am looking for a solution, like, I hope Glusterfs can fit into it, that is,
something that allows me to do live migrations of virtual machines from one
compute node to another (KVM, OpenStack).

And so I found an article about the Glusterfs-OpenStack-Connector, which was
just a basic Glusterfs setup that shared the /var/lib/nova/instances directory
across all compute nodes. This is necessary to allow virtual machines to migrate
within milliseconds from one compute node to another, as all nodes already share
every data.

Now my question is, how well does Glusterfs scale when you're having about
50+ compute nodes (on which you're about to run virtual machines). What kind
of setup would you recommend, to actually not suffer in runtime performance,
networking iops, nor in availability.

Many thanks in advance,
Christian Parpart.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://supercolony.gluster.org/pipermail/gluster-users/attachments/20120611/3126bc66/attachment.html>