The possibility exists, albeit remotely. Estrin claims Precision I/O is on track to announce products midyear. And most RDMA-over-IP components appear to be "on the drawing board."
InfiniBand proponents seem unfazed. "Powerpoints have no worth," says Stu Aaron, VP of marketing at Topspin Communications Inc., which is OEMing its InfiniBand gear to all the major server makers. The speed of InfiniBand alone ensures it will stay "a generation" ahead of Ethernet for the foreseeable future, he asserts.
"You can look at competitors in a number of different lights, but the truth is that InfiniBand remains the only standards-based clustering interconnect at the price points it offers," says Allyson Klein, a manager at Intel Corp. (Nasdaq: INTC), who also is Intel's representative to the InfiniBand Trade Association. She says HPC InfiniBand deployments number in the hundreds of thousands of nodes, and data centers are taking it up, though enterprise customers are more reticent about letting the world know they're using InfiniBand.
It's tough to predict whether InfiniBand will face a disruptive solution anytime soon. One decisive factor may be whether a new offering supports speeds greater than 10 Gbit/s, as InfiniBand now does. Both Aaron of Topspin and Klein of Intel believe that's unlikely. Even Estrin concedes that 10-Gbit/s won't likely emerge in full force until 2005 (no surprise, she says Precision I/O will be ready).
Bottom line? Despite much grousing and grumbling, InfiniBand is emerging as the next de facto high-speed clustering solution. Unlike earlier industry standards like Ethernet or TCP/IP, however, InfiniBand seems to be meeting as much resistance as market embrace. Its position could be very tenuous if an alternative actually materializes soon enough to challenge it.