Power 4 the People
IBM's RISC push
Server briefing IBM's attempt to re-capture the Unix server market continues apace. Having fallen well behind Sun and HP during the late 1990s, largely thanks to a complicated customer-confusing product line, parts of the business competing with other parts of the business and Sun's aggressive pursuit of the Internet server market, Big Blue modified its approach in October 2000. It rebranded, consolidated and streamlined its product line, and more closely allied itself to the Linux movement.
Today, a little more than 18 months on, it's hard to judge the move a success. IBM has come to dominate the Linux market, at least in terms of the money it makes. But its Q1 2002 31.5 per cent share - Gartner Dataquest's figure - arises from shipping the free OS with high ticket items like mainframes. When it comes to who's shipped the most boxes, HP holds the crown, thanks to the Compaq takeover.
And in the Unix market, IBM remains in third place, behind Sun and HP, by units and by revenue. But the economic downturn may have muddied the picture - would IBM's strategy have paid off had the server market not as far as it has?
Certainly, IBM can't be accused of not trying hard. Last October, a year on from the new strategy's launch, the company launched its latest 64-bit server processor, the 1.3GHz Power4 and quickly won plaudits for the performance of its p690 server, based on the new chip. IBM has called the Power4 a "game changer" and it's clear the chip's main role is as an UltraSparc killer.
Power4, based on the PowerPC instruction set architecture and equipped with two processor cores per die (four of which are mounted in each CPU module) has already moved into IBM's mid-range Unix servers and should be installed in low-end boxes by the end of the year, just as Sun has spread the UltraSparc III across its own product line. There are even hints that IBM is ultimately looking to migrate the technology into server blades, but clearly it has a lot of work to do on power consumption first.
IBM's aggressive roadmap doesn't stop there. Power4's successor, cunningly called Power5, is scheduled to ship in 2005 and take the chip family to 2GHz and beyond. It will incorporate Fast Path, a technology that will allow the chip to manipulate network traffic directly. Details are sparse, but Fast Path sounds like it could be some kind of SIMD technology geared toward processing IP stack data, a role typically performed in software. However the technology works, it should tie in nicely with InfiniBand high-bandwidth data links.
Power5 will also boast simultaneous multi-threading technology along the lines of Intel's Hyper-Threading. This will spilt each physical core into two or more virtual processors to handle multiple instruction threads in parallel. The jury is still out on how much of a difference this kind of approach actually makes (see Hyper-Threading score are hyper-perplexing) but there's potential for some significant performance gains here.
Finally, Power5 will bring on board elements of IBM's eLiza self-management and fault-correction technology, allowing the chip to recover from and deal with errors on the fly.
Beyond that, Power6 is expected to emerge in 2006, but it remains shrouded in mystery. Will IBM have surpassed Sun by then? Hard to say, but the technology it's bringing to Power, not to mention customer concern over just how (and when) HP is going to transition from PA-Risc/Alpha to Itanium, could yet help Big Blue move up into second place.