I’m old enough to remember when switched fabrics in Fibre Channel were pooh poohed as too costly, too complicated and not necessary – by all the arbitrated loop vendors who are now gone. My networking adolescence was spent with TCP/IP, Novell Netware, Banyan Vines, WangNet, DECNet and even some good old SNA before I got into Fibre Channel switched networking. I saw that switched fabrics had much to recommend them. In fact, they were much easier to configure and manage than the TCP/IP/Ethernet stack, scaled well, and had excellent high availability, low latency and guaranteed delivery built in. It was just what storage IO needed to overcome the problems of a parallel, copper wire, aka, the SCSI bus.
As time went by, I started thinking that a fabric architecture had wide applicability to networking in general, not just IO channels. What was interesting was the absorption of layer 3 features such as Open Shortest Path First routing, into layer 2. That fixed many problems of scaling, and link recovery since traffic can use all links between switches and loss of a link has no affect on traffic on any other link. Adding equal cost, multi-path to layer 2 with the interesting twist of switch discovery of its neighbor and learning the entire layer 2 topology, made fabric configuration dead simple. You plug a cable into two switches, wait a second for green lights on both ports, and traffic flows. Piece of cake.
Being older and greyer, I know new technologies don’t get to market unless the risk of continuing to use the “tried and true” technology is much higher. For example, when switched fabrics took off in the storage world, it was because storage growth was a force of nature that could not be resisted. Storage had to grow, very fast. The parallel copper SCSI bus couldn’t scale (in connections or distance) and arbitrated loop placed too many restrictions on topology, while root cause analysis was almost impossible when problems popped up, and problems with arbitrated loop were inevitable. The force of nature that is storage growth created way too much risk for the tried and true technology of the day, so innovation, like switched fabrics, got its day in the sun. And like a Kudzu vine, it took over storage in the data center.
Well, it’s happening again, but this time on the other side of the server, the classic Ethernet network. Switched fabrics, this time for Ethernet, not Fibre Channel, is the new innovation that is risky, but the risk of staying with the tried and true classic Ethernet tiered network relying on spanning tree has become more risky. Why is that?
Simple, another force of nature is loose in the data center, a global economic recession. Cost cutting is that irresistible force and server virtualization is an amplifier for that force. Server virtualization really cuts computing cost in a compelling way. Everyone has it, and wants more, lots more. And, another force of nature, Moore’s Law, goes along for the ride doubling the work load that can be virtualized every 18 months. Cool and inevitable.
Except, we still have tried and true classic Ethernet everywhere in the data center. And it isn’t cool. It’s the concrete wall standing in the way of those two unstoppable forces of nature. Wanna bet on who’s going to win? Yeah, it’s not classic Ethernet, that’s for sure.
The innovation that gets its day in the sun this time is the Ethernet fabric. I’m betting it’s the next Kudzu vine that takes over the data center, and this time it’s the entire data center, not just storage. All the forces of nature are aligned to ensure it.
So, that’s why you are seeing lots of articles with the word “Fabric” in them. Even the slow pokes in the industry finally get it, and wish they had “it”, a true Ethernet fabric. But, saying you have a fabric isn’t as simple as actually having one. Brocade has a long history with fabric technology. Applying it to Ethernet is inevitable. And, now we have with theBrocade VDX 6720 Data Center Switch family.