Open Stack

On Mon, Jul 28, 2014 at 4:38 PM, David Gurtner <aldavud at crimson.ch> wrote:

> Hi John
>
> Of the two ideas you describe regarding replication I believe both have
> their unique merits and would allow for some fancy uses. The main
> difference I see is in who can define what level of replications are
> available, the consumer of the API, or the operator.
> To me the second option is preferable, because the different drivers offer
> different kinds of replication (for example replication vs. erasure codes)
> and I am not sure a generic solution like the one suggested in the first
> option could offer the flexibility to optimally use the capabilities of the
> various drivers.
>
​Completely agree, and you've stated an easy summary of one of the
questions I was hoping to get input on when starting this thread.
​

> On the other hand I don't think the two options are necessarily mutually
> exclusive, and maybe the first option could exist in addition and on top of
> the second one.
>

​Agreed, I'm a big fan of iterative development, and the concept of start
small but think big :)
​

>
> At the same time I feel those are both advanced use-cases and the basic
> replication needs for a single site are already handled quite satisfactory
> by the current backend drivers. Therefore I believe the multi-site
> replication suggested by Gustavo would be the one thing where Cinder could
> really provide some added value on top of and independent of the backend
> drivers.
> Yay Pony!
>

For sure, this would be an ultimate goal.  It is sort of a Pony or Unicorn
for now, but at some point it would be the goal.  For now though it seems
independent from "which" impl to start with.

Thanks for the feedback!!
​

>
> Cheers,
> David
>
>
> On Tue, Jul 29, 2014 at 12:06 AM, John Griffith <
> john.griffith at solidfire.com> wrote:
>
>>
>>
>>
>> On Mon, Jul 28, 2014 at 4:01 PM, gustavo panizzo (gfa) <gfa at zumbi.com.ar>
>> wrote:
>>
>>>
>>> On 07/26/2014 11:28 AM, John Griffith wrote:
>>> > Hey OS-Operators,
>>> >
>>> > There are two features being worked on currently that I'd love to get
>>> > some feedback on:
>>> >
>>> > 1. Replication
>>>
>>> i would like to Cinder to be able to replicate data between different
>>> backend storages.
>>>
>>> sometimes your DR site does not have the same storage backend as your
>>> primary site, this is common when you rent the DR site.
>>> the usual workaround for this is restore from backup over the DR
>>> storage, which is painful and slow.
>>>
>>> ideally i would have a Fast dedicated storage on primary site and a
>>> bunch of lun(s) attached to a box with iscsi+lvm to provide volumes to
>>> compute nodes.
>>>
>>> if i have storage from the same 'vendor' on both sites that's great!
>>> cinder could offload data replication to vendor's appliance
>>>
>>> i know is a pony what i'm asking for, but i would really make a
>>> difference
>>>
>>>
>>>
>> ​Thanks Gustavo!
>> ​
>> ​I would really like that too, but the difficulties in doing that are
>> exactly what I don't want to introduce, at least until we get a first pass
>> with say "like storage systems" working.  Definitely understand the
>> need/desire here but at the same time I'm also not a fan of implementing a
>> feature that's just slightly more than unusable.
>>  ​
>>
>>> --
>>> 1AE0 322E B8F7 4717 BDEA BF1D 44BB 1BA7 9F6C 6333
>>>
>>>
>>
>> _______________________________________________
>> OpenStack-operators mailing list
>> OpenStack-operators at lists.openstack.org
>> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-operators
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-operators/attachments/20140728/737d4367/attachment.html>