Besides, that doesn't address the truly pitiful (when compared to Sun & IBM) throughput of the PCI bus.
Cheryl:
First, not to forget, I want to thank you for your thoughtful and clearly-expressed posts. Though we disagree on a point here and there (e.g., most of 'em :-), I respect and appreciate your integrity and thoroughness.
I think that Sun and *ALL* the rest of the PC boxmakers that will try to enter the big-server space with Itanium(*) agree with you, as does Intel. The PCI bus is a desktop bus designed around a last-generation architectural I/O paradigm. Busses are out and switches are in; hence Infiniband, whose consortium has already gotten a few revisions of a spec out, with both Sun and Intel's full participation. I haven't bothered to keep up with what the silicon schedule is for it, but everybody, Wintel chattel and Sun included, is going to abandon (at the high end) whatever bus setup they're now using for I/O and go to Infiniband within the next 2-3 years(www.infinibandta.org).
I'm sure you're aware of this development, being in the business. But in case you haven't done so yet, take a gander at the committee members listed in the spec...you'll find everybody who's anybody in addition to everybody who isn't anybody. The weakness of PCI and I/O busses in general is universally accepted and isn't relevant to the next generation of anyone's big machines. Eventually it will even replace PCI on the desktop, but that's a ways off.
--QS
*= Itanium is called out with an asterisk because there's still a chance that it will remain a boat anchor for four years and allow Sun to recover some lost ground. That chance is certainly decreasing, but it just as certainly still exists. |
