Sun Microsystems's Greg Papadopoulos shook hands and answered questions in front of the wall-sized panel of white boards he had filled with his charts and diagrams. By the end of his two-hour chat with analysts, he had put a fresh coat of paint on Sun's ages-old tag line, "The network is the computer."
That message has always been ahead of its time or expressed by people who couldn't execute on it because Sun's business model is too constraining. But for years, no matter how they dress up the distributed message, Sun has always been putting Sun-branded boxes in server rooms.
Greg's intimate chat with analysts had lots of familiar messages. On the hardware side, chip multiprocessing (CMP, in which multiple CPU units are stamped onto one circuit with a shared high-speed cache) is going to keep SPARC alive for a while. In the battle of inches that defines the CPU space, CMP buys IBM Corp., Sun, and probably Intel Corp. a few extra years for their existing architectures. But it's Papadopoulos' job to think about what happens after that. According to Sun's CTO, no existing processor can take computing into the next epoch.
Why? Because everything that's being sold now is way too expensive and too power-hungry. He grants that Intel is on top now, but he's betting that the race will be won by low-cost, high-volume chipmakers, the outfits that currently make chips for CD players, cell phones, and embedded systems. If, as he posits, Intel can't stay in business selling $500 hunks of silicon that eat 130 watts of power a piece, then neither can Sun.
He didn't stop at that. Papadopoulos said flat out that he "lives or dies" on the success of the heterogeneous, highly distributed, low-cost, and low-power computing model he described, which is quite a distance from what Sun is pushing today. Even if he was being dramatic for emphasis, he's right, and it's not just his job that hangs in the balance. It's all of Sun.
I don't know if Papadopoulos is correct that throughput is now more important than computing performance. Cisco believes that, and Sun should try extending that philosophy beyond edge devices.
However, I do think he's onto something when he says we need to apply large-scale hardware-engineering principles to distributed computing networks. Papadopoulos talks about a room full of machines as a system, and claims he doesn't care what kind of machines they are or what OS each one runs. In his plan, the era of vendor lock is over.
I understood the diagrams Papadopoulos drew. I knew what he meant when he drew a CMP processor diagram and then drew a box around it labeled "system." He was expressing some deep and important ideas, invoking CMP as a metaphor for computer networks as well as a model for building processors. I'm with him on low power, low cost, simple administration, and fast interconnects. If Sun is actually ready to follow its CTO's lead and define the leading edge instead of pretending it's already there, I'll be spending a lot of time in Menlo Park . This guy's singing my tune, and I think much of IT would love to sing along.
But if Papadopoulos' vision goes up in a puff of smoke in favor of another round of incremental tweaks to a model that's too complex, too resource-hungry, and too expensive to survive, then I'm looking to companies such as IBM, Cisco Systems Inc., Intel, Apple Computer Inc., and Texas Instruments Inc. to take us to the next big thing. It's time.