Afternoon all,
As you know I spent last week evangelising OCCI at the Cloud Computing Expo in Prague (Monday/Tuesday) and Cloud Computing Expo in London (Wednesday/Thursday), presenting the Introduction to the Open Cloud Computing Interface presentation at both. I was only scheduled for Prague but the organisers found a spot on the technical track in London too. I also ended up on the panels at both which was even more opportunities to talk about cloud interop. I'll be in Portugal for Cloud Views from Wednesday and will try to get involved in OGF 26 time permitting as well. By now people certainly know we exist and that we're doing real (hopefully good) work.
Unfortunately we're somewhat stuck on the formats decision despite hours of face to face discussion with 1/2 a dozen of the more active working group members (myself, Alexis, Chris & Richard from ElasticHosts and the Fujitsu guys). While this is not at all unusual for technical discussions we do need to fairly urgently find a solution before people (myself included) lose interest and wander off. I can't overstate how important this working group is to the future of cloud computing and both of the alternatives are rather unpalatable:
- On one side we have Amazon EC2 APIs which are not only encumbered but inelegant and inflexible (at least in the context of the enterprise use cases I spend most of my time thinking about - no offense intended). Other APIs designed to expose the functionality of a single implementation fall into the same category and while they meet their specific needs well, we need to expose the current and future functionality of all current and future implementations, not just one today. At the extreme end of the simplicity scale you have text-based APIs which we all now agree won't meet our needs.
- At the other end of the scale we have VMware's vCloud API which has been injected into the DMTF process, following in the footsteps of OVF. I fully expect the resulting "open API" to be almost word-for-word identical to the new VMware APIs which is something the DIGISTAN guys call "vendor capture". Unlike the public cloud APIs, this will be well-suited for enterprise and you can be sure that service providers will deploy VMware en masse to provide it as part of their [semi-]private "I can't believe it's not cloud" offerings. Good on them for being first and forcing the rest of the industry to follow their lead.
This working group's job is to find the middle ground - something which is simple enough to be useful for public cloud offerings but extensible enough to be useful for more challenging tasks (e.g. enterprise). This is also critical for hybrid clouds (unless you're all happy to implement DMTF's APIs in addition to your own). The business case is easily justified even if only on the basis of getting access to customers who are currently kicking the tyres with tactical deployments but unable to deploy strategically.
As you know I have been pushing Atom[Pub] hard, perhaps too hard, and the XML-xenophobes have dug their heels in as a result. It was made painfully obvious in London that blanket application of AtomPub to the problem isn't going to fly with at least one of them and to that end I've spent the weekend working on paring it back where it's not absolutely necessary. I've also purchased and read O'Reilly's RESTful Web Services book by Leonard Richardson and Sam Ruby from cover to cover and am largely sold on their concept of a Resource Oriented Architecture (ROA) - do read this sample chapter if you have time.
Fortunately I think I've found a simple, elegant solution which obviates the need for Atom (at least where collections are not required). I've captured it in a series of 3 blog posts which I'll forward to the list for the sake of convenience and the archives.
Cheers,
Sam