Of all the predictions Ive heard as people look for ITs role in economic recovery, the silliest may be No. 1 on the list of 10 released earlier this month by Gartner: "Bandwidth becomes more cost-effective than computing."
The timing of that statement is ironic since I just returned from the Microprocessor Forum, in San Jose, where the smartest people I know were talking about the difficulty of moving data between cheap, fast nodes of computation that are far more powerful than what feeds them.
Intel, for example, has an immense computing engine in its Itanium, but its a hungry monster. Fred Weber, AMDs vice president and chief technology officer, might have been comparing Itaniums instruction set to the wolf in "Little Red Riding Hood" when he warned the audience to "beware of large instruction words, the better to chew up your cache with." A megabyte of cache, Weber asserted, does as much for x86 instructions as a 3MB cache can do for long-instruction-word chips.
AMD retains the x86 instruction set, extends it to 64-bit capability and promotes scalability with HyperTransport interconnections at data rates of up to 12.8GB per second. I approve—but Id also warn that merely pumping more bits doesnt solve the problem. This is at least as much about nontechnical issues.
When someone says bandwidth is more cost-effective than computing, hes saying it makes more sense to retrieve a result than to re-create it or to get something done in one place instead of doing it at each users site. But when everyone relies on the same data and the same methods, everyone gets blindsided by the same mistakes. Thats not a flaw in our technologies; its a combination of flaws in management methods and flaws in the marketplace of ideas.
Computation needs to take place as close as possible to the points where new facts arise so that strange results receive prompt attention. People need to interact with processing so that they ask new questions in new situations, instead of force-fitting what they see to what they already know how to do.
Yes, bandwidth is getting cheap, and thats tempting, but "cost-effective" has two parts: Cut the first when you can, but dont cripple the second.
Send your predictions to firstname.lastname@example.org.