What happened to iSCSI (and InfiniBand)?

The hype surrounding iSCSI started back in 2000, when Cisco and IBM submitted to the IETF standards body a spec for running SCSI over standard TCP/IP networks. Cisco and IBM were low-key at the time, but the concept of running storage I/O over inexpensive and ubiquitous Ethernet caught fire pretty fast. Remember: At the time, Fibre Channel was suffering from severe interoperability and price problems.

Some analysts predicted that iSCSI would render Fibre Channel obsolete, and 2002 was predicted to be "The Year of iSCSI." It didn't happen.

Why? In my opinion, the key reasons were the following:

  • The economic downturn slammed vendors' R&D budgets, forcing them to put the brakes on development of new technologies. And in an economic slump, IT managers are loath to try anything new, even if it does promise lower costs.
  • The iSCSI spec entered the IETF labyrinth in 2000 and has yet to emerge as a ratified standard. I'm told that it's 99% set in stone, but final ratification isn't expected until the first quarter (and I've heard that before).
  • Early products that were compliant with "pre-standard" iSCSI specs (such as IBM's 200i disk arrays and Intel's adapter cards) fell short on performance. Compounding the problem, those products were designed to run on 1Gbps Ethernet, and they came out at a time when Fibre Channel was making the transition to 2Gbps.
  • But perhaps the main reason for the lack of iSCSI momentum has more to do with marketing than with technology or the economy. The big boys (most notably Cisco, IBM, and Intel) never really fired their sales and marketing guns. That left a bunch of start-ups holding the bag, and start-ups don't create new markets.

Is iSCSI dead? No, according to everybody. But I'm not placing bets on when it will gain traction in the end-user community. I suspect that will happen when the big guns fire.

Having said that, I think there are some end users that have started to experiment with iSCSI SANs. This past July, we planned to do a Special Report based on iSCSI users but we couldn't find any. So we rescheduled the report for the December issue. We'll see what we can dig up over the next month or two.

The prognosis for InfiniBand is even gloomier. Sure. The economy's somewhat to blame, but further hampering the fledgling technology have been recent R&D cutbacks by high-visibility players like Hewlett-Packard, Intel, Microsoft, and QLogic. And what happened to the dozens of InfiniBand-frenzied start-ups we saw lining up for easy VC money two years ago?

The fact is that the adoption of new storage protocols and technologies occurs more at the pace of a glacier than a gusher. And I should have known better:

I wrote my first article on Fibre Channel in 1994.

All together now (to the tune of Billy Joel's "We didn't start the fire"):

[1997] SCSI's dead, SSA, Fibre Channel, light the way,
[1999] SANitize, NASify, virtualize, hypnotize,
[2001] Policies, automated, SRM, over-rated,
[2003] iSCSI, IP SAN, InfiniBand, no man's LANU

We didn't start the ire...

Dave Simpson, Editor-in-Chief

This article was originally published on October 01, 2002