Cherrystone delay highlights x86 v RISC War
It's War, or something
Analysis Obscure they may be, but the skirmishes waged between the big-iron vendors' low-end RISC kit and Intel's swanky new high-end offerings get more interesting by the day.
The last couple of years have seen Intel's Xeon processors grow into a formidable armada, and while other parts of Satan Clara's execution have stumbled - pursuing the strange and disastrous Rambus fetish, or delivering late and delivering faulty - the server division has put pedal to the metal.
It's moved faster and smarter than any other part of the organisation, moving what were once overweight Pentium IIs into sleek racks, and so catching the hosting dot.com wave. It has also moved ruthlessly to quash competition from the ultra-dense blade upstarts.
Last week an elegiac Linuxgram story told us that the Compaq dream team which formed blade start-up RLX, including Gary Stimac, was taking a back seat in the operation. Coming after other business pull-outs this effectively raises the white flag from the dense-server sector. Bang goes another inflection point.
Now of all the Intel competitors still left standing, Sun is the least likely to take the Xeon onslaught lying down. And not just because it's alone in having nowhere else to go: IBM, Hewlett Packard and Compaq all have x86 and IA-64 lines that are Intel-based.
Each will have at least privately placed bets on when to let its expensive RISC design teams to go, and use Intel systems instead. In HP's case, or in Compaq's case, explicity.
But Sun has pride at stake too. Ever since the Stanford University Network formally incorporated, it's squeezed insane performance out of low-end hardware, and that engineering ethos sticks today. Heck, it has made a company philosophy out of the credo. The love of cheap fast hardware is not just a tech fetish, it is the unwritten mission statement at Sun.
A big decentralised network using hordes of ant-like computers was, in Sun's early years, the guiding vision. It chimed perfectly with McNealy's instincts - he built a pretty uniquely loose and creative organisation on these lines.
The decentralised network was also an apt metaphor for the vogue-ish techno-libertarian mores of the late-1990s, best exhibited in Kevin Kelly's Out Of Control (where command economies were like, man, brought to their knees by dispersed intelligence, y'know?).
I eat cannibals
Sun executed with its own skin-of-the-teeth improvisation: ripping up hardware roadmaps before the ink was dry, cannibalising every line faster than the competition could get to it.
This tradition took something of a back seat as Sun worked itself into a server company. A year ago, Sun was trumping the competition financially, using hardware which comfortably finished bottom of the heap on any performance comparisons.
It took a very long time - far too long - to introduce UltraSPARCIII, with no immediate ill-effects. IBM has introduced its awesome POWER4 based high-end, which gains much praise from Sun techies, and Sun has responded by refusing to take part in TPC-C benchmarks, preferring app-specific numbers instead. This looks pretty childish.
But now Sun has come out fighting.
At the MicroProcessor Forum this month Sun promised a drastic change to recent policy: higher clock speeds for a start. It promised that lots of deviant SPARC offshoots will bubble up, all competing on performance. And in the shape of Jalapeno, the cut-down UltraSPARCIII it offered the proof: an insanely-aggressive low-end processor to lick the high-end Xeon and first McKinley offerings at once.
Savannah to Burrito
The only glitch is that things don't always go according to the script. And there's been some last minute repositioning at Sun.
Sun's Daktari (up to eight-way) and Cherrystone (up to four-way) USIII servers have been talked about for 18 months now - and always in the same breath. In July Sun COO Ed Zander promised an introduction for both this fall. Well Daktari duly surfaced on Monday, in the exact form we predicted back in June.
They're large cache, low-priced servers with much of the hoopla that accompanies a mid-range launch. Good value, they certainly are: Sun spent much of the launch party in its barely-completed SoMA San Francisco office, noting price comparisons with Intel boxen, and stressing the midrangeness of the new kit.
And the figures hold true: off-the-shelf comparable x86 kit running Windows DS doesn't show any price advantages at all, and that's before you've started to tease Code Red or SirCam out of the entrails.
Sun's Neil Cox told us that Cherrystone is going swimmingly, but we should not expect a launch this calendar year. And we've heard from a couple of Reg readers (and Sun customers) who'd been looking forward to Cherrystone pricing, and were expressly disappointed.
Red Hot Chilli Peppers
Well, that's forgivable. Jalapeno systems won't ship until the middle of next year, but despite the L2 cache limitations (1MB vs the 8MB in the fully grown-up USIII) they'll comfortably match USIII performance - as they should.
Jalapeno's predecessor, USIIi, is surprisingly under-used in the current Sun range, but where it is used it makes for a dramatic marketing showstopper. This explains why expectations are so high for its successor.
We've heard from a couple of Sun customers who are prone to serious doubts as to whether Cherrystones will ever appear. They expect they'll be given the long stall until Jalapeno-based workgroup servers are ready next summer (or more likely, next fall). And can you blame them?
When we asked Neil Cox if Sun was returning to its swashbuckling 80s practice of cannibaliaing its own product lines, he gave us a grin as wide as the Cheshire Cat's. Which we take to be a "not quite yet".
The trouble with such improvisational gambits such as USIIIi, and such swashbuckling cannibalisation, is that they have to work past the suits into real products. So if anyone wants to clear up this puzzle at Senor Jalapenos or (preferably) over a pizza in Menlo Park, we'll gladly foot the bill. ®
Sponsored: Hyper-scale data management