OpenStack and Storage, a Response

In Storage, Technology by J Michel Metz4 Comments

No Fibre Channel allowed?

No Fibre Channel allowed?

During the recent OpenStack Summit in Atlanta, GA, I wrote a blog on the Cisco blog site called “Thoughts on #OpenStack and Software-Defined Storage,” arguing for the need to avoid reinventing the wheel when it comes to Software-Defined X and storage. My very good friend Stephen Foskett wrote a (fair and excellent) response, entitled Why OpenStack Doesn’t Need Fibre Channel Support.

Stephen’s argument, essentially, is that Fibre Channel, and the conventional IT that it represents, do not really have a place in the Open systems, such as the OpenStack ecosystem. Thing is, I think Stephen’s argument is a little too late. The cat has left the barn, and the genie has been let out of the toothpaste tube, the train has left the gate. Or something to that effect.

We’re Underway!

That ship has sailed!

That ship has sailed!

The issue – from an OpenStack perspective – doesn’t appear to be whether or not to include Fibre Channel, as Stephen indicates. Fibre Channel has been included (in the form of open zoning support in Havana, and zone management support in Icehouse, with additional blueprint proposals for Juno).

Instead, the question is related to the level and extent of which Fibre Channel can be managed and controlled by some type of orchestration layer using OpenStack infrastructure.

In other words, we’ve already started work with Fibre Channel in OpenStack, but the implementation is sketchy (for now). For instance, I would not recommend implementing zoning features in OpenStack at the moment due to some pretty significant bugs inside the Cinder code. This is not to say that they can’t/won’t be fixed, but obviously the work with respect to Fibre Channel is well underway.

Reinventing the Wheel

There is a key point to keep in mind as well: do we want to have a situation where we have dedicated networks to handle specific application types (e.g., such as deterministic storage applications)? We’ve spent the last six years moving from dedicated networks to converged networks specifically in anticipation of greater bandwidth capacity.

Does this mean that with OpenStack, I have to revert to 2007 designs? I certainly hope not.

At the moment OpenStack and Cinder appear to be handling the “sexy” components of DC design. Fibre Channel is not sexy, of course, but when it comes to transporting data I personally would rather have reliable over sexy any day of the week.

Having said that, I think that Stephen’s argument that other storage types (e.g., iSCSI) are “good enough” for the bulk of Data Center needs has merit, by the way. High throughput, lower latency, and greater software flexibility can, indeed, improve most Data Center applications in time.

The question for me, though, is that I thought that Software-Defined Networks (and Storage) were designed to improve Data Centers in general, not just the sexy applications, nor just for the “good enough” ones either. It appears that OpenStack, and Cinder by extension, have already decided that it’s important to manage the applications that require Fibre Channel connectivity.

That means, in turn, that if you’re going to do this, you better do it correctly.

You. Yes, you.

You. Yes, you.

At The End of the Day…

Not... quite.

Not… quite.

To that end, I actually do not necessarily disagree with Mr. Foskett (as a whole). And I’m not just saying that because he agrees with me for the most part too! 🙂

I am arguing, however, that since we have already begun to accept that applications are likely to need Fibre Channel connectivity in OpenStack environments, it’s extremely important that we do not reinvent the wheel, squeaky as it will become.

When I was in Atlanta, I saw and heard several conversations about how and what to do – questions that could be answered by asking the right people. It just so happens that those people have been doing this a long time, and belong to organizations like SNIA and FCIA.

None of this, by the way, should distract us from the key point of either Stephen’s or my original article: the relationship between storage and its corresponding network should not be ignored. Or, to put it more directly: ignore it at your own peril. There are reasons why storage networks have evolved the way they have, and to ignore those reasons because they’re not sexy enough is a significant risk that Data Centers should not – and cannot afford to – take.


  1. Pingback: Why OpenStack Doesn’t Need Fibre Channel Support

  2. It is interesting – does this mean the end of SAN A/B designs? FC has always had a higher level of reliability, but at that additional cost. I love FC/FCoE, but it will be interesting if that pricing level can survive with iSCSI, etc.

    1. Author

      I confess I’m not sure what you’re asking, actually. Can you be more specific in terms of OpenStack deployments?

  3. I know this reply is late but I don’t think it can be ingored. Was OPenstack not created to unite the industry that fails as soon as we say that using Openstack mandates certain design decisions.

    Whats next compute nodes can be anything you like as long as they are Redhat KVM. You cannot should agnostic and freedom on one end and state one choice on another.

    After all bringing in FC support is not that difficult, on the end points its almost the same and on the fabric its just a call to a zone manager to add a couple of entries. Heck it only needs updating when physical hardware is added and as such could be managed by the Bare metal provisioning system (be it manual or juju type automated).

    What I don’t think we need is huge complex FC designs being supported by Openstack.

Leave a Comment