Networking Perspectives: Nicira, VMware, and Open Source
In July, VMware acquired Nicira in what was, to many, a shocking move (or at least at a shocking price point), but in fairly quick retrospect it was understood to be exactly the right move. VMware did what any company sitting on a big pile of cash and in a strong market position would do: Look at its areas of weakness and spend the cash to acquire the technology and talent to fill in those chinks in the armor. Networking has been a rather sore point for virtualization in general and for VMware in particular. It was clear that open source projects were jumping ahead of the game (in no small part due to Nicira’s push of OpenFlow, the OpenvSwitch implementation, and the inclusion of it in the OpenStack project), and VMware, quite astutely, recognized this.
VMware, Soup to Nuts
There is no doubt that VMware plans on owning and operating the entire virtualized stack, soup to nuts. This includes compute, storage, networking and the management plane to control it all. Interestingly, one place they have not delved too deeply yet is storage; I would expect to see moves in that space in the not too distant future. In the context of this strategy, the move to acquire Nicira, the leading player in the space, was fait accompli; all that needed to be agreed upon was price. Given that Nicira knew it was in a strong leadership position, and likely had other competing suitors, they were able to command a premium. Looking at it like that, the price tag doesn’t seem quite so stupendous. Similar factors may have been in play when Cisco acquired Arrowpoint for nearly $6 billion in 2000. However, it turned out that TCP stitching was such a simple and easily improved upon load balancing mechanism that other companies were able to quickly obsolete the time-to-market advantage and become dominant players (e.g., F5, Netscaler). It remains to be seen whether a similar outcome occurs here, but it’s doubtful. Nicira’s offering is significantly more advanced and ahead of the curve than Arrowpoint was in its respective domain at a similar stage. Nicira is also a lot less about underlying technology that is subject to obsolescence and more about changing the way one approaches the lifecycle and management of networks. Anyone who has listened to the Nicira folks speak know that they aren’t technology bigots. They have always been about the larger worldview of SDN. OpenFlow and STT are means to that end; simply tools to accomplish that goal. Nicira will happily adopt VXLAN (and would have happily adopted nvGRE had Microsoft been the acquirer instead). They could even use a different flow implementation, however, given the momentum behind OpenFlow, the expectation is that the switching layer in ESXi will become OpenFlow based.
Where does this leave OpenStack?
VMware has paid lip service to supporting OpenStack, and they were (if somewhat reluctantly) allowed into the OpenStack Foundation. At the OpenStack Summit last month, VMware’s CTO indicated that VMware is committed to having vSphere support OpenStack. It would appear that VMware has recognized that a hybrid environment of hypervisors (and by association hybrid virtual switches) is inevitable in the enterprise datacenter, and they don’t want to do anything that would be an impediment to selling their value-add services and support. Of course, once the foot is in the door, I’m sure the sales team will provide pressure to migrate workloads to ESXi by touting the benefits of tighter integration with the management plane. Also, one should never forget the adage about keeping your friends close, and your enemies closer. By being part of the OpenStack Foundation, VMware gets to keep an eye on what a likely future competitor is doing (just imagine if Microsoft had gotten themselves embedded into the Linux Foundation 20 years ago).
OpenvSwitch will continue to be supported and developed, with OpenStack as its primary development platform, but perhaps with less and less involvement from the folks at Nicira/VMware over time. Even if Nicira truly believes in the project and wants to continue to wholeheartedly support it, the realities of VMware’s politics and resource constraints will cause the focus to be lost. This is not necessarily a bad thing. They gave birth to what is, by most accounts, a solid and stable product for doing virtual switching in the Linux stack; something that has been largely missing to date. It may be time to let the child go off and be nurtured less by the parents and more by the community.