Open Stack

Hi Belmiro,

In Cinder there is the "raw disk device" driver that has been used by some
for Hadoop and similar applications. That may be something to look at, but
with a big warning.

That being that it is deprecated in the Ocata release and will be removed
in Pike.

The reason it is going to be removed has a couple reasons though. One big
one is that it is unable to meet the "minimum feature set" that we require
for a Cinder driver. There are just too many restrictions on what can be
done when you're using raw disk.

The other reason was that after running a set of tests, it we really didn't
see that much of a difference in uing the raw device driver vs. configuring
the LVM driver appropriately. I don't have the full details of those tests
handy at the moment, but I'm sure we can track down some specifics if needed.

So I would recommend connecting your JBODs to some compute hosts and using
that disk space to back LVM, then configure Cinder with the LVM driver to
use that pool of space to present volumes to your compute instances.

Thanks,
Sean

On Thu, Dec 08, 2016 at 07:46:35PM +0100, Belmiro Moreira wrote:
> Hi,
> 
> we have a set of disk servers (JBOD) that we would like to integrate into
> our cloud to run   applications like Hadoop and Spark.
> 
> Using file disks for storage and a huge "/var/lib/nova" is not an option
> for these use cases so we would like to expose the local drives directly to
> the VMs as ephemeral disks.
> 
> 
> 
> Is there already something in Nova that allows this?
> 
> 
> 
> thanks,
> 
> Belmiro
> 
> CERN

> __________________________________________________________________________
> OpenStack Development Mailing List (not for usage questions)
> Unsubscribe: OpenStack-dev-request at lists.openstack.org?subject:unsubscribe
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev