Despite years of innovation and hundreds of millions of dollars invested, no networking vendor has established a leadership position in the market for data center switch fabrics.
Over the last five years, nearly every vendor has launched a new family of data center switches and fabrics. Not only do these switches offer faster bandwidth, lower latency and higher port density, they also offer a lot of innovation in the form of so-called
Data center fabrics were supposed to enable enterprises to deploy large Layer 2 networks with any-to-any connectivity for heavy server-to-server traffic in a highly virtualized data center or cloud. Cisco had TRILL-based FabricPath on its Nexus data center switches; Brocade had TRILL-based VCS on its VDX switches; Avaya embraced Shortest Path Bridging for its VENA architecture; Juniper went even deeper and offered QFabric, a solution that essentially turns an entire network into one giant switch so that Layer 2 multi-pathing wasn't even necessary.
All of these technologies have fans, but none of them have won over the market. Not even Cisco has a firm hold. The two leading analyst shops published market assessments of the data center networking industry in recent months -- Gartner's Magic Quadrant and Forrester Research's WAVE -- and neither assessment placed a vendor in the leadership position. It's pretty rare to see something like that, even if you think those market assessments are pure hype.
Juniper has taken a beating in the market for slow adoption of QFabric, but the company is a victim of setting expectations too high. Cisco may be selling thousands of Nexus switches, but how many of those customers have activated their license for FabricPath? I've spoken to countless Nexus customers, but not once has one of them told me they have FabricPath in production. If Cisco was pulling ahead in the war for fabric mindshare, I think I would have talked to a lot more FabricPath converts by now.
I have spoken to a small number of Brocade and Juniper customers who are using VCS and QFabric, but those vendors are well aware that not everyone in the market is ready for next-generation data center networking. Both vendors created something of a demarcation between their campus and data center switching lines and both are pulling back from that.
Juniper's new EX9200 is, in part, an admission that not everyone is going to adopt QFabric. Juniper is positioning the EX9200 -- which is a repackaged MX router -- as perfect for both the campus and data center core. Meanwhile, Brocade, which has had more success with its VCS fabric and VDX switches in the data center, recently announced the new ICX 6650 switch. Brocade's ICX switches are part of the company's campus portfolio, but Brocade is also recommending the 1.6 Tbps 6650 for "non-fabric data centers," as a top-of-rack or end-of-row switch.
Even Cisco's success with Nexus hasn't been able to drive out the graybeard Catalyst 6500 from customers' data centers. I've talked to network engineers who are still using the classic Catalyst 6500 -- not the newer 6500-E -- in their data centers. And they have no intention of ripping them out any time soon. These customers are just fine with their old-fashioned data center network, built in those days not so long ago when vendors sold the same switches into data centers and campuses.
Now the industry has to contend with SDN. Many SDN vendors are promising to solve some, if not all, of the problems that data center fabrics were designed to fix. Many network architects are unsure how fabrics and SDN will shake out. Do they buy one or the other? Both?
I don't think the fabric opportunity has passed these vendors by. Plenty of network architects are going to consider adopting data center fabrics in the next few years. But all of these vendors have something left to prove. The industry's poor showing with Forrester and Gartner is not a fluke. Each of these vendors needs to make sure its technology meets the evolving needs of its customers. Then each vendor needs to convince customers its technology is the best.
And what do I mean by best? Well, the solution needs to truly solve the problems of the highly virtualized data center (low-latency, east-west traffic); it has to play nice with third platforms (cloud orchestration, hypervisors, security); it has to simplify operations; it has to be well-supported by the vendor, since this is new technology; and it needs to work. Certain vendors have shipped some buggy code on very strategic core switches in recent years.
Finally, these vendors need to explain how fabrics and SDN fit together in data centers. Many of them have started doing this, but it's going to take a lot of talking, especially since the full potential of SDN is still unclear.
In the meantime, vendors need to realize that one size does not fit all. Some networking pros just won't ever need a Nexus 7000 or a full QFabric deployment. They'd rather keep using Catalyst 6500s or the EX switches, and that's fine for a company that just wants to keep Exchange and SharePoint up and running.