Interview with Sage Weil of Ceph and OpenStack
What software defined storage means for OpenStack
Recently, I had the opportunity to speak with Sage Weil, founder and chief architect of Ceph and a speaker at the upcoming OpenStack Summit in Paris.
I seized the chance to ask him a few questions about his talk and some of the things that matter most to him.
David Hurley (DH): Can you tell us a bit about yourself?
Sage Weil (SW): I started working on Ceph in 2004 as a way to solve the file system metadata scaling challenges for supercomputers, and later as a means of bringing robust, full-featured, and scalable distributed storage to the open source world and to a market dominated by expensive proprietary hardware appliances. In 2012, we started Inktank to productize Ceph and joined Red Hat in April of this year (2014). Previously, I helped start DreamHost (an LA-based web hosting company) and (for those of you who remember the web in the 90's) WebRing.
DH: That’s quite an impressive background! Congratulations on your recent move to Red Hat. Now you're speaking at OpenStack Summit this year on the topic of Software Defined Storage (SDS). What benefits do you think Software Defined Storage brings to cloud computing and in particular OpenStack?
SW: Software Defined Storage means different things to different people (mostly depending on what they are building or selling), but to me the common element is providing storage services that are independent of the hardware. For a cloud software project with hundreds of vendors representing a broad range of hardware products, there is obvious value. More specifically: once you commit to deploying a software platform that provisions and/or orchestrates cloud services across a large number of servers, you invariably need to do the same with storage. Usually, users and customers realize that platforms like OpenStack also provide them with hardware vendor independence (you can buy servers from whomever you want, usually without regard to the cloud software you are running), they look for the same sort of hardware independence in storage.
DH: It makes sense to take the same idea of deploying cloud services and apply it to storage. If users and customers are able to see the value in having hardware independence in storage solutions what do you think is the biggest problem with introducing SDS to OpenStack?
SW: The OpenStack community is dominated by existing hardware and software vendors trying to adapt their existing products to the new “cloud" messaging. For storage, this means that everyone is trying to push a definition of SDS that suits their existing products. When we were pitching Inktank to venture capitalists in 2012 and 2013 we called ourselves "software defined storage" as a joke (usually in reference to the Nicira acquisition for $1B). It was quite a surprise to find companies using the term in all seriousness a few short months later.
Today, most conversations about SDS quickly become muddled in marketspeak or definitional arguments. I find that it is more useful to use alternate terms that are more specific, like a "common orchestration API," or “open source storage virtualization," or "hardware agnostic storage software.” It is easier to have substantiative conversations about these ideas, which may or may not be included in any particular person's definition of SDS.
DH: So it’s a bit of a battleground by the various vendors all lobbying for a definition to a problem that only they can solve. It sounds like the first step is determining a common definition which everyone agrees on. Still though, even with a common definition there are some that would say the concept of SDS disagrees at a fundamental level with the very concept of an open source platform. Do you think there is conflict between the two?
SW: To the contrary, I see no conflict between the two. The common element of any SDS definition seems to be "hardware independent." One of the key values you are delivering is hardware vendor independence: you can deploy your cloud or storage platform and retain flexibility in your initial or future storage purchases without rearchitecting your solution. Deploying proprietary software, however, still locks you into a single vendor for the software, which is why I feel the true potential of software-defined anything isn't unlocked unless the software is also open source and you have a choice of vendors available to support it.
DH: That makes sense. Seems simple when you say it like that. I agree we’re not seeing the full potential of software-defined products until the software is open source and the vendor choice is open too. Ok, so I think it’s becoming apparent why this is such an important topic, but why do you think people should care about the implementation of SDS?
SW: SDS solutions all claim to provide freedom in some form. That may be freedom from specific hardware solutions, from hardware vendor lock-in, freedom from proprietary or solution-specific APIs, freedom in the sense of free software, or all of the above. Just be wary of solutions peddled as SDS that aim to do the opposite by steering you toward or locking you into a single vendor.
DH: It’s all about freedom. And like with everything we have to be careful who we trust. Ok, so back to your upcoming session at OpenStack Summit in Paris. What do you hope to see accomplished from this particular session?
SW: I was involved in a similar panel at the Hong Kong Summit last year and (somewhat inarticulately) expressed my frustration with the term and its definition. This year, I hope we can clearly identify some common ground, crisply identify where our perspectives and definitions vary, and then move on to more substantive issues, like what direction the Cinder project should take within OpenStack, and how it should position itself relative to other so-called SDS solutions on the market.
DH: That sounds awesome. It sounds like over the past year you’ve been able to really refine your thoughts on the subject. I’m quite sure everyone attending will come away with a much better understanding and appreciation for SDS. I have really appreciated you taking the time to give me (and our readers) a sneak peek about what you’ll be sharing. This has really been an informative meeting! Before I go are there any final thoughts you’d like to share?
SW: The storage industry has long been dominated by hardware appliances bundled with proprietary software. n almost all cases, that hardware is really commodity components wrapped in branded tin, with the true capabilities of the system determined by the software "secret cause.” The true industry revolution comes when you get the storage functionality and performance you need with open source software that will run on any hardware you choose. Then you get the transformative collaborative innovation you saw in the OS space with Linux over the last two decades and the resulting user freedoms and lower costs. Common orchestration APIs are great, but if it's just a new way to talk to the same entrenched players that have been fleecing you for years then you're missing the real opportunity.
DH: That’s a great takeaway. It’s important to look at all aspects of software defined solutions and ensure openness. Well, let me close by saying we’ll certainly be looking forward to hearing more and I am sure the session at OpenStack Summit will be amazing! Thanks for taking the time to talk with me.