InfiniBand has long suffered from the problem of having to serve two masters at once. On the one hand, it excels at high-end switching environments. On the other, its cost and limited industry support make it an expensive proposition.
And now that Ethernet is starting to push the 40 G and 100 G levels, InfiniBand is having an even tougher time justifying its utility.
Which is why it comes as a surprise that Intel would shell out $125 million for QLogic's InfiniBand business. Ostensibly, the company hopes to use the technology to drive high-performance computing (HPC) platforms into the ExaScale universe, but the question remains whether InfiniBand is still the best way to go about it.
While it's true that Fourteen Data Rate (FDR) InfiniBand offers nearly twice the bandwidth of QDR versions, the latency remains the same. So on the one hand it supports scaled up operations, but does nothing to accommodate the need for greater speed being driven by solid-state storage environments and increasingly sophisticated multicore processors. It's important to note, however, that QLogic still has its FDR products under wraps, so perhaps Intel has already seen something that at least makes the investment worthwhile.
Of course, where market leaders tread, others tend to follow. So it will be interesting to see if other top-tier vendors start to show a similar interest in InfiniBand. As The Register's Timothy Prickett Morgan points out, that could place a number of suitors at Mellanox's door, possibly even AMD, which could use both the SwitchX and ConnectX-3 platforms as key components to its processors and/or chipsets.
But as I said, just because InfiniBand makes technological sense does not mean it makes business sense. Research firm Bishop & Associates reports that InfiniBand is likely to overtake Fibre Channel in the overall networking market, but both will continue to trail behind Ethernet by wide margins for the foreseeable future. With volume shipments unlikely to happen, does InfiniBand offer enough of a value proposition to continue as a niche solution, or even as a component on a converged fabric platform?
Intel clearly thinks so. It will be interesting to see how it intends to make it happen.