Back in the 1990s when the original PCI I/O bus was getting a bit long in the tooth, two disparate groups of vendors proposed solutions to the problem. Compaq, IBM and Hewlett-Packard championed a standard called Future I/O, while Intel, Microsoft and Sun Microsystems pushed a competing technology called Next Generation I/O. In an unusual act of solidarity, the two groups got together, compromised on their differences, and came up with a jointly developed technology called Infiniband.
In simple terms, Infiniband is a switched I/O channel that connects processors to other processors and high-speed peripherals like disk drives. The best industry analogs are mainframe channel technologies ESCON and FICON.
When it was first proposed, Infiniband seemed like a much better solution than either Ethernet or Fibre Channel. Infiniband proponents claimed that Gigabit Ethernet was too slow and TCP/IP too processor-intensive for computer operations and I/O. As for Fibre Channel, the thought was that it would die because of pure economics. As companies like Cisco, HP and Nortel produced tons of Infiniband devices and millions of ports, Infiniband price/performance would leave Fibre channel in the dust.
Fast-forward to today and Infiniband has to be seen as a mere footnote in the annals of technology. Yes, I realize that the technology exists and is deployed in lots of high-performance computing environments, but rather than become an industry staple, Infiniband is an extremely esoteric technology. In the meantime, high-density 10Gb Ethernet switches from Extreme, Foundry, Force 10 and, of course, Cisco are becoming staples in today's virtual data centers serving SOA and Web 2.0 applications, while fast processors and TCP off-load chips eliminate all the fuss over chatty and slow IP packet processing.
I'm sure someone will respond to this blog by telling me that Infiniband is far more prevalent than this and that I am a moron for suggesting otherwise. Before someone does this however, I suggest that they scan through the past three months of CNET articles and see how many times they can find a reference to Infiniband. Few, if any.
I can't help but think that Infiniband may be the next ATM, Token Ring or Betamax--a superior technology that never gained broad market penetration. I just wish I had 5 percent of the money spent on Infiniband development, technology evangelism and industry hyperbole.