Sequent Computer Systems

Sequent Computer Systems, or more commonly just Sequent, was a computer company that designed and manufactured parallel computing systems. Together with Pyramid Technology, Encore Computer, Alliant Computer Systems, and AT&T Computer Systems, they were pioneers in symmetric multiprocessing (SMP), inventing algorithms such as read-copy-update. Through a close partnership with Oracle, Sequent became a dominant high-end UNIX platform in the late 80s and early 90s. Later, after several missteps, they returned to their roots, producing a next-generation high-end platform for UNIX and Windows NT based on a non-uniform memory access (NUMA) architecture, NUMA-Q. As hardware prices fell in the late 1990s Sequent found their market shrinking, and eventually they were purchased by IBM in 1999. Although the acquisition was made to establish NUMA-Q-based systems as the high end of their Intel-based platform line, changes in senior IBM management led to changes in strategy, and the death knell for NUMA-Q was sounded when in 2002, two layoffs at Sequent's former headquarters in Beaverton, Oregon ended all development on the systems for which IBM had acquired the company.

History

Sequent formed in 1983 when a group of eighteen engineers and executives left Intel after the failed iAPX 432 mainframe on a chip project was cancelled. They started Sequent to develop a line of SMP computers, then considered one of the up-and-coming fields in computer design. Several engineers from AT&T Bell Labs also came over, bringing systems programming expertise.

Sequent's first computer systems were the Balance 8000 and Balance 21000 released in 1984. The Balance included up to 20 National Semiconductor NS32016 processors, each with a small cache connected to a common memory to form a shared memory system. The systems ran a modified version of BSD Unix they called DYNIX, for DYNamic unIX. The machines were designed to compete with the DEC VAX 11/780, with each of their inexpensive processors dedicated to a particular process. In addition the system included a series of libraries that could be used by programmers to develop applications that could use more than one processor at a time. The Balance systems were originally intended to be sold to OEMs as computing engines, but that market could not be developed. When the commercial market discovered their reliability and cost advantages, the company re-thought its marketing strategy. The Balance line sold well for three years to banks, the government, other commercial enterprises, and universities interested in parallel computing.

Their next series was the Intel 80386-based Symmetry, released in 1987. Various models supported between 2 and 30 processors, using a new copyback cache and a wider 64-bit memory bus. 1991's Symmetry 2000 added SCSI drives, and were offered in versions with 1 to 6 Intel 80486 processors. The next year they added the VMEbus based Symmetry 2000/x50 with faster CPUs.

The late 1980s and early 1990s saw big changes on the software side for Sequent. DYNIX was replaced by DYNIX/ptx, which was based on AT&T's version of UNIX instead of BSD. And this was during a period when Sequent's high-end systems became particularly successful due to a close working relationship with Oracle, specifically their high-end database servers. In 1993 they added the Symmetry 2000/x90 along with their ptx/Cluster software, which supported various high availability features in general and Oracle Parallel Server in particular.

In 1994 Sequent introduced the Symmetry 5000 series models SE20, SE60 and SE90, which used 66Mhz Pentium CPUs in systems from 2 to 30 processors. The next year they expanded that with the SE30/70/100 linup using 100MHz Pentiums, and then in 1996 with the SE40/80/120 with 166MHz Pentiums. With the addition of a VGA card and the Winserver NT software, the 5000 series could also run Windows NT.

Recognizing the increase in competition for SMP systems after having been early adopters of the architecture, Sequent sought its next source of differentiation. They licensed their technology to Intel to help commoditize the SMP market, and began investing in the development of a system based on a cache-coherent non-uniform memory architecture (ccNUMA). NUMA dedicates separate portions of memory to different processors, avoiding the bottleneck that occurs because only one processor can access memory at a time. Using NUMA would allow their multiprocessor machines to generally outperform SMP systems, at least when the tasks are tightly coupled with their memory -- as is the case for servers, where each user tends to be looking at different files.

In 1996 they released the first of a new series of machines based on this new architecture. Known internally as STiNG, an abbreviation for Sequent: The Next Generation (with Intel inside), it was productized as NUMA-Q and was the last of the systems released before the company was purchased by IBM for over $800 million. IBM then started Project Monterey with SCO, intending to produce a NUMA-capable standardized Unix running on IA-32, IA-64 and POWER and PowerPC platforms. This project later fell through as both IBM and SCO turned to the Linux market, but is the basis for "the new SCO"'s SCO v. IBM Linux lawsuit.

In 2002, after Sun Microsystems began a public discussion of IBM's silence on their NUMA-based x430 system, IBM had a reduction-in-force, and announced that it had no further plans to market the x430 and would eventually drop support for the over 10,000 systems that Sequent and IBM had deployed.