The first Open vSwitch (OVS) “conference” was held over the phone, and the second was a small affair hosted by Cisco last year. So when Ben Pfaff, a VMware principal engineer, came up with the idea of hosting a more formal conference, it didn’t seem like so big a deal.
Organizers were even skittish to reserve an auditorium at VMware‘s Palo Alto, Calif., headquarters, because a reservation required at least 100 attendees. They didn’t know if they’d make it. But by the time the Open vSwitch 2014 Fall Conference began, the estimated crowd of about 40 had turned out to be a full 200, with a waiting list capped at 100.
Held Nov. 17 and 18, the conference still had the feel of a clubby gathering. Attendees were free to wander into a VMware kitchenette for snacks — and we were asked to replenish the coffee ourselves whenever the pot ran out. (And I did.)
OVS has some heft. It’s the switch provided with the KVM and Xen hypervisors, and it’s grown to be the favored switch in OpenStack, used by 43 percent of production deployments and 49 percent of development and quality assurance environments, said Guido Appenzeller, CTO of VMware’s networking and security business unit, during his conference talk.
That beats out the Linux bridge — the switching option built into Linux long ago — which is in 15 percent of deployments.
But two years ago, SDxCentral raised a crucial question: Is it healthy for an open source project to be under the curation of a single company? Nicira was essentially running OVS and deciding which improvements would make it into the code, and Nicira is now owned by VMware.
Talking to people during and after the conference, though, the consensus seems to be that the former Nicira team is making an honest effort to keep OVS open. A couple of competitors even remarked that they’ve been pleasantly surprised at the results. And companies including Cisco and Intel seem enthusiastic to contribute to the cause, although Intel did try creating a fork of OVS at first.
Open vSwitch at VMware
Pfaff was OVS’s first committer: “He locked himself in a room for a few hours and came out with Open vSwitch,” says Justin Pettit, a VMware staff engineer and OVS’ second committer.
That was at Nicira, and from there, Pfaff’s colleagues began contributing to OVS. “The bulk of the commits do come from VMware, but we’ve worked the longest on it,” Pettit says.
That actually makes sense. OVS isn’t that old, and there hasn’t been much time for the Nicira team to disperse to other companies.
It’s true, though, that VMware is stingy about who gets to be a committer — that is, who gets to actually alter the OVS source code. “They won’t give me committer access,” Appenzeller told me with a smile. (Appenzeller joined VMware just recently from Big Switch Networks.)
To become a committer, you have to make contributions and review code, both in substantial amounts, and then get nominated and then win a vote. “Being an employee of VMware or Nicira is not enough,” Pettit says.
Pettit doesn’t say that to boast about VMware’s pickiness, though. He’s actually making the opposite point: That the OVS project is willing to make committers out of non-VMware contributors. It just takes time.
Pettit also denies that VMware is cherry-picking which code submissions get approved or reviewed. “The only reason code hasn’t gone in is because it’s buggy or the architecture wasn’t right,” Pettit says. Code that would sink the performance of OVS, for instance, doesn’t get a review.
Being picky does have its down side, of course. The performance of OVS has been an issue for a while. The subsequent fixes, leading to what Rackspace called “ludicrous” performance, were the highlight of one of VMware’s few talks at the conference. But the situation points to what could be an ongoing PR problem for OVS: Any time things aren’t going swimmingly, users are likely to bring up VMware’s concentration of OVS brain power.
The issue could be less about trust and more about practicality.
“I think, had Nicira continued to be independent, they would have continued to put more effort and weight behind OVS, and with the VMware acquisition, clearly this slowed down substantially,” writes Recep Ozdag, a director of marketing at Cyan, in an email to SDxCentral. Lack of openness isn’t a problem with OVS, he writes. His bigger concern is that OVS continues to progress at a reasonable clip.
Any slowness perceived in OVS comes from a conscious tradeoff. The OVS team are being sticklers about the quality of code, which necessarily means not going a full throttle.
“It’s not as bad as some people claim. Feature velocity is good, and [the core committers] want to make sure that there is stability,” writes Dan Conde, director of product for Midokura, in an email to SDxCentral.
OVS in the Long Term
So far, the OVS team does have some separation from the rest of VMware. Pettit heads a team of about a dozen that’s trying to keep functioning as an isolated unit, theoretically keeping VMware’s corporate fingerprints off of OVS’ planning.
It’s not clear whether VMware would be interested in curating OVS in the long term. The answer would really depend on whether the Nicira team stays. If they scatter to the winds during the next few years, then there’s a good chance they would continue contributing to OVS under their new company banners. If OVS became that widespread an effort, it might make more sense for VMware to hand the project over to an entity such as the Linux Foundation, or to a newly created foundation.
For now, the community seems satisfied that VMware is curating OVS in good faith, and VMware put a lot of effort into making sure the OVS conference wasn’t VMware-dominated.
“We are committed to keep Open vSwitch open. We have no illusion about what this means,” Appenzeller said during his conference talk. “There are some people selling against us with OVS, and that’s OK.”
(Photo: The Open vSwitch conference in Palo Alto, during a break. The auditorium was standing room only during the talks.)