VMWorld has started and in the opening keynotes it was mentioned that the Nicira deal had closed late last week, which meant the acquisition was fair game for discussion. At the end of the first day there was an Industry Analyst Q&A with Gelsinger and Maritz, and I got a question in about Nicira. I’ve known Pat since his Intel days, and I know he enjoys direct questions, so I asked him why VMware had spent so much for Nicira. I granted that it was only 3% of the VMW market cap, and that SDN is important and Nicira a first-class team, but on the other hand VMware already had an excellent and OpenFlow knowledgeable network team, but viewed another way, $1.2B is a lot of money (twice what EMC paid for VMware, and more than Peet’s coffee was acquired for more or less at the same time.) “So where,” I asked Pat, “is the meat?”
Pat smiled and started to answer but Paul grabbed the mike (it was his deal after all). He started by reminding us that in 2002 companies thought that $100M was too much to pay for VMware, and today that is a $40B mistake. He then spoke passionately of the profound impact of server virtualization on IT (Amen) and suggested that the impact of network virtualization could be similar, in which case $1.2B for the best of breed could well turn out to be a similarly smart decision. In the keynotes Nicira had been presented as helping in VMware accelerate federation plans though their contribution to the OpenStack network stack. But that by itself clearly doesn’t justify the price. In the Q&A Paul talked additionally about the fact that as CPU’s increase in cores more and more VM’s will share a physical network attachment, and more and more of the inter-VM traffic will necessarily be handled in the virtualization layer (not the physical network). The implication was that the Nicira technology would be valuable here although no specific functionality was presented. That makes more sense. Then, in finishing his answer to me, Paul talked about distributed SDN controllers and how difficult it is to develop them, and that is an area where the Nicira team excels.
SDN controllers are indeed an important and under discussed topic. In the simple view of OpenFlow, all the complexity of distributed switches is replaced by a single shared controller. That might be possible for research efforts but practically the issues are much more complex because controllers have to be replicated for performance and availability, and need to be moved close to the switch hardware in order to solve performance problems. So a practical and reliable controller for a large network will be distributed and complicated, and without a doubt, SDN controller distribution is a critically important problem, although once you understand that fact, it diminishes the apparent advantage over conventional networks because you’re replacing one form of distributed complexity (the distributed control plane of a legacy network) with another form of distributed complexity (the distributed controller).
So I guess the simple answer is that Paul spent 3% of the market cap as insurance against reiterating the mistake of not buying VMware for $100M in 2002. In the past, Nicira has been very calculatedly secretive about exactly what they have built so it’s hard to fully appreciate what this brings to the VMware network offering. More will come out in time I’m sure.