Sun plots critical path to post-SMP future

Systems companies such as Sun Microsystems could do worse than draw on Buckminster Fuller for inspiration for their future. At an analyst briefing in Menlo Park last week, Greg Papadopoulos said that Sun's challenge was navigating a "post-SMP" era of computing, which is a bold statement from a company that mushroomed in size on its ability to do big SMP systems convincingly.

N1 is the first stage of Sun's response, but unkind wags have suggested that the "N" stands for "nebulous" or "nothing there yet".

Sun isn't alone in needing to convince people that "systems" have a future. Here's where Bucky could help.

Sky map

Fuller had been studying Eratosthenes intriguing map of the world, draw c.200BC (CE.) and was convinced that Phoenician sailors had circumnavigated the world.

The map is certainly a grand puzzle. The "Northern Coast" of what appears to be the Euroasian continent is defined in some detail. Strangely, there's a substantial gulf there in the center of this landmass. But the continent is icebound in the north, and was only navigated much later, by nuclear submarines. There's an Elizabethan version of the Eratosthenes' map, which interprets this gulf as the Caspian Sea.

But Fuller wasn't buying this. What was needed, he decided, was a new projection. Fuller's Dymaxion map provided an explanation. Lacking compasses or propulsion, the sailors followed the currents. They turned left at the Straits of Gibraltar, rounded Cape Horn and followed the Indian Ocean current west and then the Japan currents North, skirting the Bering Straight. Currents took them down the West coast of the Americas, to cut a long story short, the Gulf Stream took the home. All of which is highly implausible if you insist on using the projection we learn at school. "Using Mercator's, the polyconics, or other world projections, it would never have been logically revealed," he wrote in Critical Path. (There is other evidence, we now know, but it's exceptionally difficult to believe the route on such a map).

Fuller's projection doesn't move any continents, it simply redraws the map: and that's the moral of this tale. The Phoenician daubings on the coast of South America might be the requirement for people who demand "empirical" proof. But to get people to really believe something, sometimes you just need to find a new way of describing what they already know.

Papadopoulos and his software CTO John Fowler were attempting to dispel a couple of ideas. The technical proposition, that better systems logically followed bigger and faster processors, and the business challenge that Sun would be able to provide such systems efficiently, thereby avoiding the fate of a Wang or a DEC.

Beware Greeks bearing GIFs

Gordon Bell, said Papadopoulos, had suggested that computing is ripe for change every ten years. Which for Sun meant looking at a post-SMP world.

Papadopoulos acknowledged that post-UltraSPARC V, even Sun will find itself designing systems differently. He said to expect core densities higher than what we see today.

"Itanium, UltraSPARC V, and POWER are not the way chips will be designed in the future," said Greg.

"When you get to 4, 8 or 16 processor on a chip; then you start to a see a system on a chip where the performance is really outstanding."

It was absurd to build systems around processors that dissipate 135W per chip. Nor did higher frequencies do anything to solve the memory latencies. The fastest-clocked processors in the world were still waiting around for something to do.

The first "post-SMP" or "network scaled" systems would be - and this is surely too tautological to be useful here - "the first systems that are built from networks, rather than attach to them."

Papadopoulos agreed that much of the "bitmass" of data remained on notebooks and isolated storage systems, but Sun vies with IBM to produce systems that draw this to a virtualized center. Microsoft has its own approach to pooling data, of course.

Asked what Sun's sales proposition might be in three years' time, Papadopoulos said it would be TCO: ownership costs were "obscene" and Sun needed to prove managers could install a turnkey system rather than have to budget 4-8 weeks production time to get a system up and running.

Great Leap Forward

The perceived wisdom on Wall Street is that the great systems companies won't be able to survive commoditization: which means Sun, and SGI, and those parts of HP and IBM which consider themselves systems companies - as opposed to being call centers or credit companies (HP and IBM are both, too ) - must fall to the logic of horizontal economics.

"We're always be under margin pressure," Fowler told us later. "But the current level is sustainable."

But one analyst present (forgive us, we didn't catch your name) took this idea to its natural conclusion. No company which had a vested interested in a particular component - such as a microprocessor - should deserve the title "systems company", he suggested.

An original proposition, and the chap conceded that such a company probably doesn't exist right now.

But it does. Step forward, Michael Dell - by this learned definition, you run the only "true" systems company in the world! ®