DefCore isn’t a new creature in the OpenStack community: it’s been in discussion since at least 2013. However it was only earlier in 2015 that adherence to DefCore Guidelines became a requirement for products that want to use the OpenStack name and OpenStack Powered logo. As a result, a lot more people are now interested in DefCore than in the past. As we’ve started receiving a lot more feedback about DefCore, it’s become apparent that there are a few misconceptions out there from folks who are new to DefCore. Personally I held some of those same misconceptions when I started participating in DefCore about 9 months ago. This series of posts is intended to lay some of those misconceptions to rest and help folks learn more about DefCore is, how it works, and what it’s trying to accomplish.
One item of feedback we’ve received a couple of times now is that some folks expect the DefCore Committee to be a visionary body that sets forward-looking standards that others will be expected to adhere to. In essence: DefCore defines what makes a cloud interoperable, and then everyone else gets to work implementing those capabilities in their products.
This actually isn’t the case (today).
Well, more accurately: it’s only partly true. DefCore does set standards that vendors have to adhere to if they want to use the OpenStack trademark or logo…but most of how those standards are defined is by looking at what the market has already accepted.
The criteria that the Board of Directors has approved for DefCore to use in establishing interoperability standards today are strongly weighted toward requiring capabilities that have already gained wide adoption–not what we’d like the world start adopting. In essence, DefCore’s Guidelines are trailing indicators of the acceptance of a feature. The evidence of this lies in the 12 Criteria that DefCore uses to decide which Capabilities become required in it’s Guidelines. Currently, all 12 Criteria are weighted roughly equally. If we examine them to see whether they’re more of a leading indicator or a trailing indicator, we find:
- Widely Deployed: trailing. No points are awarded for this Criteria unless products are already supporting the Capability in question.
- Used by Tools: trailing. Tools are mostly unlikely adopt a new feature/API the minute it’s released. Rather, they tend to lag and add support for it when the tool’s users demand for it is sufficient.
- Used by Clients: trailing. See Used by Tools.
- Future Direction: leading. If the project (for example, Nova) or the TC says a Capability is going away in the future, no points are awarded even if the Capability isn’t gone yet.
- Complete: trailing (sort of). If the capability is plugin dependent, all plugins have to support it. That doesn’t always happen immediately when a feature is released (unless the project requires it) so we’ll call that a trailing indicator.
- Stable: trailing. The capability has to have been present in 3 or more OpenStack releases, which means the capability has to have been released for 18 months before it gets points here.
- Foundational: trailing. Other capabilities are unlikely to be dependent on a very new thing. In fact there are numerous examples of the opposite: for instance, until around the Juno timeframe neutronclient (on which any Neutron capabilities that require authentication would depend) supported only the Keystone v2 API. The Keystone v3 API was introduced as far back as Grizzly.
- Atomic: neither.
- Proximity: neither.
- Discoverable: neither.
- Documented: neither (or maybe in some instances trailing). In most cases major new capabilities get documented when released (the community has generally gotten a lot better about this over the years IMHO). There are still some cases where a capability isn’t well documented up front though, and does get documented at some later point. In that sense you might call this a trailing indicator.
- DefCore in Last Release: trailing. A capability only scores points here if it was required six months ago in the last DefCore Guideline as well (which also means the Capability was likely present in at least three OpenStack releases per the “Stable” criteria, so in essence the Capability is at least two years old if it gets points here).
So of the 12 criteria, 7 are looking at trailing indicators of adoption, 4 aren’t really either, and 1 is a leading indicator. It’s also maybe worth pointing out that using current levels of real-world interoperability as the starting point has been discussed by longtime Board members since at least 2013, so the idea that DefCore is weighted toward trailing indicators of adoption isn’t new. That said, adherence to DefCore’s Guidelines only became a requirement to get a trademark/logo license agreement from the OpenStack Foundation in the past 6 months or so, so we probably now have a lot more people now paying attention and interested in providing feedback.
Some of those discussions have indicated that at least part of the community feels that the current criteria are perhaps too heavily weighted toward trailing indicators. Personally, I tend to agree. To that end, some discussions have already begun about proposing changes to the Board of Directors concerning weighting of criteria.
Questions? Comments? Drop by #openstack-defcore on IRC or drop the DefCore Committee a line at email@example.com.