[ovirt-users] Posix FS as alternative to local storage?

[Eduardo Mayoral]

On 30/04/18 12:51, Tony Brian Albers wrote:
> On 30/04/18 11:43, Eduardo Mayoral wrote:
>> Hi,
>>
>>      I would like to set up a new oVirt deployment with hosts that have
>> the VMs running on local attached storage. I understand this has the
>> requirement of having each host in its own cluster (and own datacenter,
>> it seems, I understand the need for the dedicated cluster, not so much
>> for the dedicated datacenter).
>>
>>      At the same time, I would like to have some shared storage domains
>> so I can use it to export VMs or migrate them around hosts (probably in
>> three stages, first migrate VM storage from local to the shared storage
>> domain, second migrate the host (probably not possible to do a "hot"
>> migration, but at least "cold"), third migrate the VM storage from the
>> shared storage domain to the local storage domain of the new host).
>>
>>      So I thought maybe I can deploy a datacenter in shared storage mode,
>> with one cluster per host. Use one or two shared storage domains for
>> master and as an stage area for planned VM migrations as explained
>> before, and then configure several storage domains, one per host, as
>> posix FS . I would then deploy the VMs on the local posix FS storage
>> domains and set affinity rules for the VMs to their hosts as needed.
>>
>>      Would this work? Is there a better way of achieving local storage
>> and retaining the ability to share storage among hosts and migrate VMs?
>>
>>
> Have you thought about using glusterfs? If hosts are physically close, 
> that would probably be the best solution.
>

Actually, yes, I also had glusterfs in mind. However one of the main
reasons to use local storage is performance, and I am concerned about
the write latencies of gluster (If using gluster, I would handle things
so the VM runs on one of the gluster nodes hosting the VM data, so I
assume the read latency will be close to the one I would get with local
storage, but the gluster replica(s) will be on other hosts, so write
latency may be significantly worse).

Thanks a lot for the suggestion, it is a good one, however, the original
question stands: Would this work? Is there a better way of achieving
local storage and retaining the ability to share storage among hosts and
migrate VMs?

Best regards,

--
Eduardo Mayoral.