Open Stack

I suspect that the original poster was looking for instance access
(mediated in some way) to IB gear. When we were trying to figure out
how to best use our IB gear inside of openstack, we decided that it
was too risky to try exposing IB at the verbs layer to instances
directly, since the security model doesn't appear to have a good way
to prevent administrative commands from being issued from untrusted
instances.

We decided to use to IB as fast plumbing for data movement (using
IPoIB) and have ended up with pretty nice I/O performance to the
volume service, etc. We haven't managed to use it for much more than
that at this point.
 -nld

On Mon, Oct 3, 2011 at 11:44 AM, Caitlin Bestler
<Caitlin.Bestler at nexenta.com> wrote:
> Nick Khamis asked:
>
>> I was wondering if any effort has been made regarding supporting
> infiniband
>>devices to support I/O Virtualization, RDMA etc..? If so, can you
> please direct
>>me to the latest documentation.
>
>
> Adding RDMA support to OpenStack will be a challenge with the current
> software architecture.
>
> What is really needed is adding zero-copy I/O to eventlet socket I/O.
> Explicit RDMA semantics
> being visible to OpenStack code would only be possible if *every* device
> in a cluster supported
> RDMA, and that would happen so seldom that the code taking explicit
> advantage of RDMA would
> Become a maintenance liability.
>
> But refactoring the eventlet sockets I/O to use I/O virtualization when
> available is distinctly possible.
> It would be major work, but would greatly reduce the memory bandwidth
> consumption. What you
> would need to do is convince one of the high speed NIC vendors (10 GbE
> or InfiniBand) that it was
> worthwhile to invest in some Python libraries.
>
>
> _______________________________________________
> Mailing list: https://launchpad.net/~openstack
> Post to     : openstack at lists.launchpad.net
> Unsubscribe : https://launchpad.net/~openstack
> More help   : https://help.launchpad.net/ListHelp
>