Published: 11 Aug 2008
The IT industry can be like an accordion, playing v-e-r-y s-l-o-w–l-y. After eight or 10 years of gradual expansion, there's a period of consolidation that seems to take nearly as long. I'm not sure what the IT accordion sounds like, but I think vendors must hear sweet music as we purchase new technologies and squeeze out old ones.
A few decades ago, IT was strictly centralized. Big iron was the name of the game, and the centerpieces of corporate data centers were massive mainframes hooked into dedicated disk and tape systems. If you wanted to do anything computer related, you got on the waiting list to claim a slice of mainframe time.
That didn't work all that well, especially when PCs popped up and people realized that instead of queuing up for the mainframe they might be able to get some work done. Feeling threatened, IT declared the mainframe dead and set about building distributed computing environments with networks of shared storage.
Fast forward to today. Consolidation is in, and silos or islands (whatever you want to call them) are out. Where 10 servers were once scattered, there's now a single dual-quad core hosting a dozen virtual machines. That doesn't sound very distributed, but the legerdemain of server virtualization makes things feel distributed. Gee, that sounds like partitioning a mainframe ... and whoosh!, the IT accordion squeezes some of the air out of the server environment.
Storage isn't quite as easy to squeeze, however. Virtualization helps, but not in the same way it does for servers. Server virtualization makes one big server look like a bunch of servers, while storage virtualization makes a bunch of storage systems look like one giant array. Although it seems a little inverted, the concept is the same because it pulls all those disparate arrays into a single management framework. It's a control thing, which isn't bad, I guess, even if everything is beginning to look kind of mainframe-ish again.
Next up? The network or, to be more precise, networks. This is the big pipe theory embraced by Brocade and Cisco, but in different ways. Fibre Channel (FC) was created because Ethernet was good at what it was good at, but not so great for storage networks, particularly when it came to performance.
But now we have 10Gb Ethernet and even a reworking of the venerable protocol called Datacenter Ethernet that promises speed, reliability and low latency to rival FC. Wrapping everything in a tidy package without leaving legacy FC behind is Fibre Channel over Ethernet (FCoE), which is probably the most popular acronym being tossed around storage shops these days. FCoE is the wunderkind of networking. It can carry TCP/IP and FC data, and at a speed that exceeds the current top ends for both protocols. And it could get faster.
Cisco seems to think FCoE is the best thing since sliced bread, or at least lost packets, and it's betting its one big pipe in the data center on the technology. Brocade doesn't appear to be quite as convinced, but it has embraced FCoE, just not as tightly as Cisco. No surprise here: Cisco, the master of all things Ethernet, sees FCoE as the future; Brocade, with its feet firmly in the FC world, says FCoE is nice enough, but it's more an extension of FC than a replacement.
But these networking giants do share a common view of one massive network serving as the data center's digital spinal column. Hey, isn't that what InfiniBand is supposed to do? With InfiniBand, you can have a high-speed backbone that supports FC and Ethernet along with native InfiniBand protocols. Everyone likes new and cool, so FCoE gets all the attention, while InfiniBand has been lurking in the wings forever, it seems. Maybe it's time to take another look at InfiniBand.