Prediction: Fast is not everything

The Internet Engineering Task Force has about finished the first set of the IP storage standards, so you should start to see iSCSI and Fibre-Channel-over-IP products soon, at least by some definition of “soon.”

IP storage might become an almost-perfect case study for Clayton Christensen to use in a follow-up to his book The Innovator’s Dilemma because it will so clearly show how hard it is for people already in a business to properly understand the important features of a disruptive technology. I predict that 1) this technology will be very successful, and 2) the main success will be in just the area many professional storage people dismiss as uninteresting.

The idea behind the IETF’s IP storage protocols is quite simple. Just encapsulate SCSI, which is used to connect small disk drives to PCs, and Fibre Channel, which is used to connect big disks to big computers in data centers, into an IP-based transport protocol. See the IETF IP Storage Working Group Web page for more information.

There were two areas that generated major angst in the early work of this working group. (Not to imply that the working group is now an angst-free zone, but it’s better than it was.) The first area was security. When the working group charter was approved, it specifically required that all implementations of the IETF IP storage protocols had to include strong security (for cryptographic data integrity and confidentiality). Users do not need to use them if they do not want to, but the ability to turn these on must be in the product before a vendor can say its product meets the standard. Quite a few working group participants really did not like this requirement – they figured that the main use of these protocols would be in a data center or some other area protected by a firewall. But once you put an application on IP there is no way for the application to be sure where it is being used, for example behind a firewall. This is a major feature of IP.

The second area is performance. A number of people in the working group and the analyst community are quite focused on making sure that the IP storage protocols can run very fast because disk drives are very fast these days. Who would want a slow drive? I expect that the implementations will operate at a high speed. Tests have shown that the Macintosh laptop I have can transfer data at over 450M bit/sec (it has built-in Gigabit Ethernet); I would expect it will be able to run IP storage protocols at nearly that speed.

But the biggest thing IP storage has going for it is the flexibility of IP, and that performance is a secondary issue. In the Christensen book, slower, smaller disk drives won the market over bigger, faster disk drives. The same thing will happen here, and the vast majority of IP storage use will be at low speed.

Disclaimer: Harvard has had a long time to figure out how to do things slowly, but this prediction of the importance of flexibility is my own.