When InfiniBand was first announced, it was seen as a potential solution to many different problems. For some, it solved the PC/server bus bottleneck, being seen as a replacement for the legacy PCI standard. For others, they saw InfiniBand as a server peripheral interconnect, possibly becoming a new SAN networking technology. Some even saw it going even further and eventually replacing Ethernet as a LAN protocol.
This is, of course, all normal behavior, and similar responses have been seen with the introduction of almost all new networking technologies. I remember seeing similar claims for Fibre Channel just a few years ago, boldly proclaiming that we would soon (by the year 2000) see Fibre Channel connections to the enterprise desktop and that TCP/IP on Fibre Channel would eventually replace Ethernet as the networking technology of choice. And before Fibre Channel, the new all-encompassing technology was seen as ATM. Before that was FDDI, and so on.
What has typically happened is that, for a variety of reasons, new technologies like InfiniBand get placed in a single "killer" application area. This helps the people marketing the technology to better target their initial efforts and helps customers differentiate between potential solutions. For Fibre Channel, this was, of course, storage area networks. For InfiniBand, it appears it will be server interconnects. Already, quite a number of blade server vendors, including CTX and Compaq, are using it in this way. In fact, BlueArc will be using the technology in this fashion in the future.
As for the future of InfiniBand, very few new technologies get to break out from their niche. Even the king of networking, Ethernet, has always been seen as a LAN transport and is (with new fiber optic-based Gigabit Ethernet) just starting to explore new markets in the MAN and WAN spaces. It seems likely that the same niche play will be true for InfiniBand, at least for the time being.
Copyright 2002, Blue Arc Corporation.
This was first published in June 2002