With the quarterly reporting season almost upon us, analysts at Merrill Lynch & Co have been casting an eye across the current server computing landscape and despite vendors’ best attempts to blind us with science and performance numbers, it’s hardly a surprise the brokerage’s most recent survey of corporate buying behavior finds users don’t buy benchmarks, they buy a relationship. Nevertheless it’s apparent, the analysts say, that sustained differences in product performance, however, can affect vendor financial results, and in both directions if you are Digital Equipment Corp. While Merrill Lynch figures that Sun Microsystems Inc and Hewlett-Packard Co will continue to gain share in the high-end Unix servers it’s got a downer on expectations for both companies’ March quarter revenues, yet puts IBM Corp on the upside. While HP reclaimed the price-performance lead in February with its V class systems, thanks to its use of fewer processors to do the same work, Sun will likely regain the lead when it publishes TPC-C results for its high-end Starfire running the 64-bit Solaris 2.7 in the third quarter. IBM scored absolute top marks with its SP 309 system at 57,054 tpmC but needed 12 nodes with 8 processors each to do so. Indeed IBM’s numbers are reckoned to be gradually improving which, combined with its server sales force now representing the entire product line, could result in improved server sales later this year. Moreover, software is only 12% of total costs for HP’s V2250 compared to 32% for IBM’s SP 309 and 34% for Sun’s 6000 Cluster. Also, it observes, the single instance of Sybase on the V2250 cost far less than multiple instances of Oracle Parallel Server required on the Sun and IBM cluster systems.

Unix not yet in peril; no NUMA mindshare

NUMA architecture, as represented by Sequent Computer Systems Inc, Silicon Graphics Inc and Data General Corp isn’t yet a big success: mindshare is low and performance has not leapfrogged the big boys. It certainly hasn’t been able to displace the predominant position of symmetric multiprocessing solutions. Windows NT is not a high-end play and Unix still also predominates in the mid-range although NT owns low-end servers and leads the midrange in price-performance. Moreover NT lags Unix in virtually every operating system dimension except cost and PC interoperability. We doubt NT 5.0 will be the Unix-killer; more likely it’s NT 6.0. That will give Unix at least another couple of years’ grace we suppose. Importantly it looks as if HP’s ability to integrate NT and Unix could yield long-term advantages relative to Sun’s Unix-only posture. Merrill says five of the top six midrange servers by price-performance were Pentium Pros running Microsoft SQL Server and although Unix systems offered the best performance in the midrange: Wintel’s attack from below should result in Unix share loss and pricing pressure the next few years. The low-end market (under 14,000 TPC-C) was increasingly dominated by PC servers. The Unix workstation market declined last year for the first time as PC workstations have moved to the top of midrange workstation graphics performance pile. Worst affected was SGI which has lost most of its performance advantage; its systems cost three times that of its rivals. Merrill believes Sun will have increasing difficulty growing workstation sales through share consolidation.

Some chips may fall by the wayside

While IBM’s benchmark numbers are showing gradual improvement, mainframes pitched for server consolidations don’t often beat open system platforms. Its Parallel Sysplex mainframe clustering technology delivers increased performance, high availability, and software cost savings, but requires rewriting applications and lacks support of some third-party software vendors. IBM tells the brokerage that 40% to 45% of its customers use Parallel Sysplex with a higher proportion enabled to use it. Merrill thinks the AS/400e is now more competitive, the RS/6000 SMP narrowing its disadvantage, while the SP2 is at the top of the Unix performance heap. Market researcher IDC finds AIX 4.3 to be the leading platform for delivering 64-bit commercial Unix functions ahead of DEC and HP. Merrill’s dire predictions for the microprocessor market: the chip is no longer of strategic importance is sure to fuel more of the Sun dumps Sparc for Intel type musings. Merrill itself says, we could see Sun dumping SPARC for Intel but maintaining its commitment to the more strategic Solaris operating system. And MIPS is hardly used as a computer processor anymore. While RISC still has performance advantages – Alpha and PA-RISC are the top commercial processors while Alpha and MIPS are best for technical applications – with the advent of Merced, we expect a number of architectures to fall by the wayside, perhaps including PA-RISC, Sparc, MIPS, and Alpha, it says. Its view of benchmarking is common sense 101. They often do not include most product configurations from smaller companies due to the high costs of benchmarking. It’s clear rankings can change often as companies submit new benchmark results using systems with faster processors and optimized tuning and some benchmarks may be based on systems and software releases not yet available. Users tell Merrill that system reliability, vendor reputation, and service capability are as important as performance and price considerations. Strategy and mindshare are more important than performance figures.