By William Fellows
IBM Corp has agreed to pay $810m cash or $18 a share for its Unix partner Sequent Computer Systems Inc whose ccNUMA technology it will use to build distributed shared memory into the entire range of Big Blue servers. The price being paid is a small premium over yesterday’s $17.18 close but Sequent’s stock has already rocketed over 25% to $17.44 when news of an impending deal was leaked last week. The market judged the acquisition no big deal and IBM closed at $137.06, down $0.31.
In throwing its hat into the ccNUMA ring IBM will give new momentum to those companies which believe that to achieve a new level of scalability beyond symmetric multiprocessing, processors must be able access pools of distributed shared memory rather than simply directly attached RAM. Hewlett-Packard, Silicon Graphics Inc and Data General Corp are among those already treading that path. The ccNUMA penalty is the time it takes, or latency, to reach those distributed resources, and is the reason why Sun Microsystems Inc’s public policy is to decry ccNUMA as a speed bump and keep its foot firmly on the SMP gas. IBM server chief Bob Stephenson says that ccNUMA will be a defining technology for Unix servers in the first century of the new millenium.
IBM has created a ccNUMA division under Tony Befi which will operate out of the RS/6000 facility in Austin, Texas, charged with developing ccNUMA technologies for deployment across all IBM servers, AS/400 and S/390 included. Code, hardware and ideas are all in the pot. Eight Sequent engineers are already working in the group which Befi says will combine Sequent’s ccNUMA technologies with IBM’s existing ccNUMA work and deliver it in different forms across the platfoms. IBM wouldn’t say why it had decided to pay Sequent $810m for ccNUMA rather than building out its own technologies. It’s known to have been working on a ccNUMA for other AS/400s since 1991 and has been working on a ccNUMA for an RS/6000 project called S-COMA for some years. ccNUMA will follow clustering on to AS/400. ccNUMA’s scaling techniques – not distributed shared memory itself – will be applied to S/390. IBM says it will develop a new version of S/390’s Parallel Sysplex which will employ ccNUMA techniques for more tightly integrating CPUs. It also believes it will be able to build certain ccNUMA functions right on to the chips it uses in its servers.
Some of Sequent’s ccNUMA technology is built into its Dynix/ptx Unix and that functionality is already being integrated into the Monterey Unix being co-developed by IBM, Santa Cruz Operation Inc and Sequent for future Intel IA-64 systems. Based upon an Intel port of IBM’s RISC-based AIX Unix, Monterey will also include pieces of SCO technology and SCO will also resell Monterey as its 64-bit Unix for Intel. IBM says the acquisition of Sequent will enable it to accelerate the Monterey project.
Sequent’s Intel-based server line will extend IBM’s eight-way NetFinity series of Intel servers into the enterprise. Sequent currently supports 64 CPU, with expansion to 256 in the pipeline. IBM says it remains totally committed to its PowerPC-based RS/6000 servers which are also sold into the high-end commercial Unix server market. It claims 80% to 90% of customers are self- selecting in any case.
IBM says that Sequent’s ability to co-host Windows NT on its servers was the second reason behind ccNUMA that it was attracted to the Beaverton, Oregon company. Actually Sequent uses this capability more to get a foot in the door than anything else. There’s little if any revenue on it but the capability is sure to see greater demand as NT reaches up into the enterprise and deploy a mix of applications. Dynix’s ability to dynamically partition a system into different virtual systems was the third hook, IBM explains. Some of that work also makes it feasible to run Unix and NT under one hood. It will be building these functions into Monterey and other system software.
Why Sequent CEO Casey Powell has decided to sell now – beyond the Unix and Intel synergies the two companies have been flagging wildly – isn’t clear. To be sure Sequent has been stymied from growing past the $1bn barrier by its size, and companies buying such high-ticket servers upon which their businesses depend want to see something big and solid standing behind it. IBM certainly brings that. Powell says Sequent will now be able to flex IBM’s muscles. It will also be able to attract a raft of ISVs, the like of which Sequent has not thus far been able to tap. Sequent gets access to IBM’s sales channel – both IBM and Sequent sales forces will sell Sequent servers, initially under the existing NUMAcenter and NUMA-Q brands. IBM has said it will expand its own server sales force accordingly. Ultimately the Sequent groups will be absorbed into IBM divisions.
Sequent’s share price declined steadily to $5.69 in October last year from a high of around $30 in mid-1997 when the first wave of optimism surrounding ccNUMA broke. Moreover Sequent has always described its real opportunity as the IA-64 market, which is now only months away. Perhaps it knows something we don’t. Perhaps Intel’s Merced, or even McKinley, the second generation IA-64 that most enterprise vendors will target, are going to be later than currently thought.
We wonder whether part of the reason IBM has taken out Sequent is to prevent it falling to Gateway, Dell or another PC maker looking to get into the enterprise. With Sequent’s ccNUMA rival Data General Corp now valued at only $885m they wonder if Dell can afford not to buy it. After all Dell already gets its high- end Fibre Channel storage from Data General. Data General might make an uncomfortable meal because of its legacy systems baggage and but CEO Ron Skates has so far refused to break the company up, for sale or as an ongoing businesses.
