[Openstack-operators] Shared Storage for compute nodes

Jonathan D. Proulx jon at csail.mit.edu
Tue Jun 21 14:51:28 UTC 2016


On Tue, Jun 21, 2016 at 11:42:45AM +0200, Michael Stang wrote:
:I think I have asked my question not correctly, it is not for the cinder
:backend, I meant the shared storage for the instances which is shared by the
:compute nodes. Or can cinder also be used for this? Sorry if I ask stupid
:questions, OpenStack is still new for me ;-)


We use Ceph RBD for:

Nova ephemeral storage
Cinder Volume storage
Glance Image storage

(and ceph for object storage too)

/var/lib/nova which holds the libvirt xml files that actually define
instances live on local node staorage.

This is sufficient for us to do live migration.  However as of Kilo at
least 'vacating' a failed node doesn't work as it assumes
/var/lib/nova is on shared storage if the ephemeral storage is shared
even though the xml could be recreated from the database.  I don't
know if Juno or Mitaka still have this issue or not.

If I were trying to solve that I'd probably go with NFS for
/var/lib/nova as it's easy, storage is small (just text files) and
load is light.  But we've been very happy with ceph rbd for ephemeral
storage.

Our use case is private cloud w/ 80 hypervisors and about 1k running VMs
supported by a team of two (each of whom has other responsibilites as
well).  Ceph is 3 monitors and 9 storage nodes with 370T raw storage (
with triple replication net storage is 1/3 of that.)

-Jon

: 
:Regards,
:Michael
: 
:
:> Matt Jarvis <matt.jarvis at datacentred.co.uk> hat am 21. Juni 2016 um 10:21
:> geschrieben:
:> 
:>  If you look at the user survey (
:> https://www.openstack.org/user-survey/survey-2016-q1/landing ) you can see
:> what the current landscape looks like in terms of deployments. Ceph is by far
:> the most commonly used storage backend for Cinder. 
:> 
:>  On 21 June 2016 at 08:27, Michael Stang <michael.stang at dhbw-mannheim.de
:> mailto:michael.stang at dhbw-mannheim.de > wrote:
:>    > >    Hi,
:> >     
:> >    I wonder what is the recommendation for a shared storage for the compute
:> > nodes? At the moment we are using an iSCSI device which is served to all
:> > compute nodes with multipath, the filesystem is OCFS2. But this makes it a
:> > little unflexible in my opinion, because you have to decide how many compute
:> > nodes you will have in the future.
:> >     
:> >    So is there any suggestion which kind of shared storage to use for the
:> > compute nodes and what filesystem?
:> >     
:> >    Thanky,
:> >    Michael
:> >     
:> > 
:> >    _______________________________________________
:> >    OpenStack-operators mailing list
:> >    OpenStack-operators at lists.openstack.org
:> > mailto:OpenStack-operators at lists.openstack.org
:> >    http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
:> >  > 
:>  DataCentred Limited registered in England and Wales no. 05611763
:> 
:
: 
:Viele Grüße
:
:Michael Stang
:Laboringenieur, Dipl. Inf. (FH)
:
:Duale Hochschule Baden-Württemberg Mannheim
:Baden-Wuerttemberg Cooperative State University Mannheim
:ZeMath Zentrum für mathematisch-naturwissenschaftliches Basiswissen
:Fachbereich Informatik, Fakultät Technik
:Coblitzallee 1-9
:68163 Mannheim
:
:Tel.: +49 (0)621 4105 - 1367
:michael.stang at dhbw-mannheim.de
:http://www.dhbw-mannheim.de

:_______________________________________________
:OpenStack-operators mailing list
:OpenStack-operators at lists.openstack.org
:http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators


-- 



More information about the OpenStack-operators mailing list