VMworld Has VXLAN Winning the Network Overlay War
August 29, 2012
While less than riveting--or, in VMware's case, less than live--there were enough VXLAN demos from vendors like Arista, Avaya, Brocade, Cisco, VMware and NEC to be noteworthy. There was some expectation that VMware would make a bigger splash with Nicira, but the deal closed Thursday before the event--so expectations may have been overblown. The lack of a stunning Nicira demo aside, it's clear that VXLAN is gaining traction, and that's a good thing.
VXLAN isn't necessarily the best option in the battle for overlay dominance, but it's a reasonable option. If there is agreement on VXLAN, perhaps the competing protocols that are sprouting like weeds will die before the market is littered with them. Let's look at what we have. There are three network overlay protocols aiming for standardization: VXLAN, NVGRE and STT. Each has a different set of backers.
- Are You Really Ready for a Software Audit?
- Want Information Fast or Want it Right? Learn How to Have Both
White PapersMore >>
Then there are two protocols to perform edge virtual bridging in hardware: 802.1qbg virtual Ethernet port aggregation and 802.1qbh VN-TAG. There are also two protocols undergoing standards development for doing multi-path Ethernet--TRILL and Shortest Path Bridging--plus three proprietary multi-path Ethernet protocols already shipping: Brocade VCS, Cisco FabricPath and HP's IRF. Yes, all three really are proprietary.
There's also OpenFlow, which changes how Ethernet paths are defined in hardware. Finally, there are two proprietary protocols for Wide Area Ethernet, Cisco's Overlay Transport Virtualization and HP's Ethernet Virtual Interconnect.
That's 13 protocols, all aimed at modernizing Ethernet. With the exception of the last two, 11 are focused on virtual networking and easing operational problems with virtual machines and networking.
What enterprises and service providers need is a single overlay standard that everyone from hypervisor vendors to hardware vendors can implement. Once that's in hand, the networking industry as a whole can move onto more interesting problems like programmability, tenant isolation and L4-L7 service insertion. If the number of demos I saw at VMworld is an indication, VXLAN is the leading candidate.
Some of the VXLAN demos were proofs of concept, like Arista showing one top-of-rack switch acting as a tunnel endpoint, and Avaya, which showed VXLAN traffic moving across an SPB multipath Ethernet network. There was also Brocade, which showed its ADX as a tunnel endpoint and load balancer. VMware didn't have a demo. It had a canned animation showing how VXLAN would work. The trend was unmistakable: VXLAN is picking up some momentum--we are seeing shipping code now with more products coming by the end of the year. Given VMware's market presence and its support for VXLAN, I can only see demand growing.
Faced with the reality that the market is signaling its choice, maybe the IETF's Network Virtual Overlays working group can agree on the initial documents and move toward a standard protocol. But don't expect a miracle; I doubt we'll see any viable standards for at least three years.
VXLAN is here, and it's being implemented in vSphere and hardware. I think that trend will continue and more vendors will get on board. Microsoft can try to hold out for NVGRE, but unlike in years past where its domination in Windows Servers could dictate protocols like MS-CHAP and SMB, it's far behind VMware in hypervisor market share. Given VMware's push and the networking vendor's pull, it looks to me like VXLAN is the winner.