Software is the Hard Part
July 27, 2012 5 Comments
Boy, that escalated quickly
Well, that was unexpected. The Super-Duper big news Monday (July 23rd) is that VMware has purchased Nicira, and it seems (at least right now in the heat of the moment) that this is a game changer for the industry. I’m writing down my thoughts on this subject here, which may or may not turn out to be insightful. I’m just spitballing here. For some great perspectives check out Brad Casemore’s post and Greg Ferro’s post on Network Computing.
First, this could mark a turning point when networking moves from mostly a hardware game to mostly a software game. With Nicira, VMware has purchased arguably the most advanced control plane out there by a long shot. And unlike Cisco’s Insieme spin-in and Cisco OnePK, Nicira is a shipping, mature(-ish?) product with big-name customers (though how widely deployed it is unknown).
Most are in agreement that this change from hardware to software was coming, but I think that most (including myself) figured this change would take a few more years. Nicira had been an interesting curiosity until now; a few big name customers sure, but not really market penetration. With VMware having an presence in just about every data center in the world, Nicira can be pitched/adapted to a much, much wider audience.
Why the change from hardware to software? Mostly the commoditization of network hardware. Vendors like Broadcom and Intel (through the Fulcrum purchase) offer SoC (Switch on Chips) and other Ethernet silicon that can be (relatively) easily engineered into a switch with much less R&D than was previously possible. As has been mentioned, most of the network vendors use these now to build their switches. Cisco has been pretty much the lone hold out in this trend, continuing to invest in their own R&D, chipsets, and hardware. Even Juniper’s ambitious QFabric is believed to run on top of Broadcom Trident+ chips.
This will be a challenge not just for Cisco, but Juniper, Arista, etc., as the differentiation in capabilities and performance between their silicon versus commodity silicon is declining. If you can’t differentiate in hardware, they’ll have to create new differentiations.
Nicira could potentially take away one of those differentiations: Software. Cisco, Juniper, Brocade, and others have been working on software differentiations. Cisco has several technologies, including FabricPath (very cool, but licensed badly), Juniper has QFabric, Brocade with VCS Fabric, etc. Cisco also has the Nexus 1000v, and while Nicira is not an immediate threat, it could potentially put a wrench in those gears.
Nicira is a very advanced control plane, in many ways very different than anything currently out there. Most people run the usual suspects for their control planes: Spanning-tree, OSPF, BGP, etc. Slightly more modern control planes are TRILL, SPB, VXLAN and NVGRE, OpenFlow, and Juniper’s QFabric. But none of them tie it all together end-to-end quite like Nicria does. And now it belongs to VMware. And because it belongs to VMware, Nicira now has exposure into almost every data center on the planet.
In a Nicira SDN world, the only thing you need from a network vendor in a data center is a network built with OSPF and inexpensive Broadcom/Fulcrum-based switches. You wouldn’t need TRILL/SPB, FabricPath, QFabric, VDS, or even spanning-tree (since every pair of switches would be its own Layer 3 domain). The Nicira/SDN controller would create the overlay based on whatever overlay network technologies (VXLAN/NVGRE/STT) between the virtual switches located in the hypervisor.
Espcially with an SDN-dominated world, there’s not much to differentiate on hardware anymore, and in an SDN world software won’t be much of a differentiation either, since one vendor’s OSPF isn’t going to be different than another’s.
In terms of SDN, Cisco, Juniper, et al are behind. For starters, neither Cisco nor Juniper have really lead the SDN charge. They’ve both opened up or announced that they will shortly open up their switches and routers with APIs that will allow SDN controllers to control them, but they both lack a controller of their own and have seemed for the most part to have more of a defensive strategy against SDN (since it potentially distrupts them).
This could be tough for Juniper, Cisco, Arista, etc. to overcome. They’re mostly geared as hardware companies. Turning them into full-fledged software companies will be a challenge. Insime is supposed to be SDN related, but who knows if it’ll be an answer to Nicira, or something more anemic.
As with all of this, time will tell. Things are changing so fast, it’s impossible to predict the future. But one thing I am fairly certain of is that software, as Nokia and RIM have figured out, software is the hard part.