IBM, the B-52 of computing, has launched a mainframe cruise missile at the rapidly-growing market for Unix and Windows NT servers. Its new S/390 Multiprise 2000 is aimed at medium-sized businesses -- and new customers in emerging markets like China, India, or Eastern Europe -- who can be persuaded to desire the good old blue virtues like reliability in a low-priced all-in-one package. At the same time, IBM announced that it has jacked up the power of its CMOS processors, reportedly doubling the performance of its previous S/390 SMP (symmetrical multiprocessing) units in the largest 10-processor configuration. IBM has also given the upper-end upgrade a name that is twice as long: IBM S/390 Parallel Enterprise Server -- Generation 3 (hereafter Gen-3). The Multiprise and Gen-3 use the same basic CMOS technology. The Multiprise is, however, capped at five processors. It also comes with a luncheon special of system and application software, maintenance, internal disk drive option, and support services aimed at a reduced-fuss startup and low prices. Both new product lines will be available for general purchase at the beginning of October. THE RINGMASTER'S WHIP In contrast, the Gen-3 scales up to 10 processors in each SMP unit. More important, it features IBM's Parallel Sysplex tiered clustering system, which uses the SMP units as nodes, scaling to giant compute factories with up to 320 processors. The Sysplex is not your ordinary string-and-tin-can cluster. It operates under the whip of a distinctive ringmaster that allows direct, concurrent read/write access to shared data from all the processing nodes in the configuration. A workload manager provides dynamic workload balancing based on customer-determined business objectives. As an IBM white paper puts it, "The Parallel Sysplex approach eliminates bottlenecks, ensuring a consistent response to users since work is distributed across the configuration and all the servers can get to all the data." The ringmaster of the Gen-3 and earlier CMOS machines in the S/390 product line is a Sysplex timer. Until now, large-scale clusters required that the entire herd of SMP nodes be in the same room or very nearby. The Sysplex timers had to be close to one another because interconnection was limited to a few meters of copper cable. This limit has been extended to three kilometers through the use of optical cables. This provides more flexibility for customers who use their Gen-3 nodes in a distributed mode around a large installation or corporate campus. IBM argues that clustering is sufficient to meet the high-capacity needs of the most demanding customer. NETWORKED, OF COURSE Indeed, IBM is placing growing emphasis on the use of mainframe systems in a networked setting. New cryptographic hardware and other features are described as qualifying the S/390 products as ideal servers for intranet and Internet applications. IBM is also seeking customers who drank the client-server medicine but found that it tasted sour and, even worse, cost more over the long run. In Bluespeak, this is "reintegration." The networked philosophy is retained, but an S/390 becomes the "integration technology" (i.e., new server). GEN-4 IN 1997: EQUALING BIPOLAR IBM prefers to give the impression that the Gen 3 version of the S/390 family is the final word in mainframe-style computing. It isn't that simple. CMOS technology provides advantages that IBM repeats again and again: lower cost, lower operating expenses (due to air-cooling and a much smaller footprint) and higher reliability. (The advantages of less-fragile CMOS chips are enhanced by the distributed architecture of the S/390, especially in the Sysplex mode, which reduces the risk of a full-system crash if one component goes bad.) Nevertheless, modern CMOS technology (IBM currently works at device separations of 0.5 micron, which is competitive but not super-advanced) is not yet quite as powerful, on a per-processor basis, as the latest versions of the bipolar technology that was the foundation of mainframes for so many years. IBM hopes to attain the goal of equaling -- or surpassing -- bipolar single-processor performance with a promised Gen-4 product, expected in 1997. Meanwhile, IBM has stopped development of bipolar technology. (Enough bipolar components will still be produced to support customers wishing upgrades.) MAINFRAME WARS CONTINUE This opens a window for Japanese vendors who have, for many years, competed successfully in the plug-compatible market. They are offering new parallel products as well as mainframes based on bipolar technology. Hitachi has in fact developed a new chip that combines both technologies. For the next year or two, at least, Hitachi and the other plug-compatible competitor, Amdahl/Fujitsu, should continue to do quite well, responding to customers who want lots of power but fear (or disdain) IBM's clustering. Very high profit margins have traditionally been one of the strongest allures of the mainframe strategy. The switch to CMOS and market competition from client-server and other solutions (including IBM's own RS/6000 SP distributed memory machines) have forced IBM to lower prices. Thus, although the total volume measured in MIPS is unprecedently high, IBM's mainframe revenues have been halved since 1990. Japanese vendors are joining the lower-price parade. RUNNING FASTER TO KEEP UP Thus, IBM must run faster and faster to avoid further reductions in revenue. This is not just a question of hardware sales. The mainframe franchise is really the MVS franchise -- the proprietary operating system that IBM has nurtured for decades. This led in turn to a sturdy, very profitable market for IBM software running on MVS. In order to keep up with the times, IBM has renamed MVS as OS/390. It has provided for linkages with Unix and Windows NT. It reports proudly that independent software vendors (ISVs) will have ported 1500 applications to the S/390 by the end of 1996 -- with 200 of them provided by ISVs that have been Unix specialists. IBM has already paddled through one patch of rapids that could have wrecked the whole enterprise. It has made sweeping technological changes in the foundation of its mainframe business while convincing its customers (or enough of them) that, in a spiritual sense, that are stilling getting the same familiar old products. Nevertheless, IBM's corporate health could decline seriously if the MVS franchise deserted the old blue flag. The Gen-3 upgrades and a number of simultaneous announcements of new products in mass storage and other fields are intended to shore up the loyalty of its MVS customers. At the same time, IBM is deploying the Multiprise junior mainframe in attempt to broaden the franchise among customers with less money and shorter patience with the costly complexity of traditional mainframes. This amounts to a basically defensive strategy. Although it involves a good deal of technological creativity, it is not as glamorous as the forms of parallelism that have turned their backs resolutely upon the mainframe tradition. Yet, in American football as in some other sports, a very aggressive defense can take the ball away from the opposing team and score goals -- sometimes enough to win the game. IBM hopes to do the same. Unlike games like football, however, the rules of today's market in commercial computing are not fixed. They change every day. IBM, like its competitors, will not find it easy to keep up. Additional information about the S/390 can be obtained from IBM's S/390 Web site http://www.s390.ibm.com ---------------------------- Norris Parker Smith is a journalist who specializes in HPC and high bandwidth communications. Reader comments are welcome.
IBM’S NEW CMOS JUNIOR MAINFRAME TARGETS UNIX AND NT MARKETS
September 13, 1996