NCR maps out the revised growth path for its Teradata users as it unveils enhancements to the NCR 3600…
When NCR Corp bought Teradata Corp back in 1992, users of the El Segundo, California firm’s line of DBC/1012 back-end database engines were promised a migration route to NCR’s planned 3700 system. The 3700 was seen as the next step up from both NCR’s 100-CPU, high-end systems 3600 and the Teradata machine. With Teradata technology at its core, the 3700 was originally to have been released at the beginning of this year. By last year that date had slipped to 1994, and earlier this summer NCR advised that the 3600 should provide Teradata users with their upgrade path. The 3700, a Pentium-based machine with thousands of CPUs compared to the 3600’s maximum of 100 processors – is now conceived as a package of incremental enhancements to the 3600, not a system per se. At last week’s launch (CI No 2,293), Teradata founder Philip Neches, now senior vice-president and chief scientist at NCR, admitted the company had changed its plan for the 3700. He said features that should have gone into the 3700 and become available as a box swap for Teradata and 3600 users are instead being incorporated into the 3600. Andre Dahan, vice-president of NCR’s large computer systems northeast division, said the firm originally thought it could fork lift users from the 3600 to the 3700, then realised it would have to do something else and decided upon a gradual course of upgrades. The NCR 3600, which has shipped since April, now has the Teradata database, Oracle, Ingres, Informix and Adabas databases up on it – whereas the DBC/1012 machine comes with just the Teradata option. Beyond the 3600 Pentium upgrades for the DBC/1012 announced last week, Teradata users are promised continued support, limited upgrades and release 1.5.1 of the Teradata database through to 1995. Although revenue from sales of the two parallel lines has been roughly equal since April and DBC/1012s are still being delivered – Neches says all new customers are being steered to the 3600. Following the latest round of enhancements, 3600 users can expect a new raft of additional technologies (3600 Release 3) around 1995, including higher speed Ynet interconnects, Release 1.5.1 of the Teradata database, Pentium Access Module Processors, 3.5 disk storage technology, enhanced system, communications and peripherals management and some new middleware components. 3600 Release 4, due in 1995 or beyond, will include Release 1.6 of the Teradata database, Oracle 8.0 and Asynchronous Transfer Mode communications. By this timeframe, NCR will have its 3700 technology in place so that 3600 Release 4 and 3700 Release 1 are effectively common nodes. Pentium architecture, Teradata for Unix, Sybase and Oracle databases will be common to the lines, but the 3700 will incorporate NCR’s next-generation BYnet interconnect system. NCR’s existing Ynet tree-type interconnect enables two inputs and one output at any node. BYnet is regarded as a banyan or folded tree system that accommodates two inputs and two outputs, enabling the same number of nodes to be employed at the top as at the bottom of the architecture. BYnet will deliver the bigger bandwidths required for multimedia and other types of processing.
NCR’s brand of parallelism lies in the MIMD – multiple
instruction multiple data – technology camp. MasPar Computer Corp and Thinking Machines Inc use an alternative SIMD – single instruction multiple data – form to deliver parallel processing. Within the MIMD camp there are shared memory and shared nothing architectures. IBM Corp’s parallel systems are based upon the use of shared memory, a concept implemented in software, which Neches doesn’t believe will scale easily to new hardware requirements. Shared nothing constructs are used in nCube Corp and Intel Corp parallel systems, as well as by NCR. It is at the interconnect that these systems finally differ – Intel and NCube preferring a mesh-type system of joining nodes with NCR using tree and folded-tree mechanisms. Despite a growing band
of hopefuls in the world of massively parallel commercial systems ICL Plc, NCube, IBM Corp and Meiko Scientific Ltd to mention a few – Neches says they are all delivering unproven, first generation technologies. Moreover, he believes, many are tying pricing structures for their systems too closely to the world of mainframe residuals, where terms are based on methods of discounting from a price customers would otherwise pay for similar mainframe class performance. Neches says NCR pricing is now based firmly on the CPU performance curve, not the mainframe performance curve, a trend he believes will see traditional mainframe residual pricing collapse.
Parallel market leader – and investing heavily in it
NCR, which is building a $60m facility in San Diego to manufacture its parallel systems, currently invests around 20% of its research and development budget in parallel technology development. Teradata shipped its first back-end database engine in 1983 and together with the 3600s installed since April NCR has some 400 parallel systems installed at 180 sites – 50 of those systems are 3600s running Oracle parallel server. Most recently, the Reuters news agency has ordered a 3,500 MIPS 3600 system – the largest parallel system in Europe, Neches claims while US retailer WalMart Inc with a 476-processor DBC/1012 handling 3Tb of data is the largest commercial parallel user in the world, NCR says. With relational database technology only now delivering the robustness associated with traditional systems, Neches says there is little incentive, from NCR’s point of view, to put newer object-oriented technologies up on these machines yet, although it is tracking developments, he says. The company commands the lion’s share of the parallel market, and dominates the commercial sector. Smaby Group puts NCR’s revenue from massively parallel systems at some $260m in 1992, from a market worth around $520m in total. Thinking Machines weighed in with $91m; Intel, $73m; MasPar $26m, nCube $16m, Meiko $16m and others $17m. The market, which includes scientific and engineering as well as commercial systems, is expected to grow at around 20% a year. The difference between the two is that in commercial processing 90% of the work is done at the system software level, while up to 99% of work undertaken by scientific and engineering systems takes place within the application. That’s why NCR’s work on developing parallel extensions to Unix System V.4 is so important, argues Neches, extensions that he would eventually like to be taken up a standard components of a common Unix application programming interface set.
NCR enterprise systems get a new customer focus
Apart from its specific parallel announcements, NCR has also introduced a new customer focused business model for its enterprise systems market, under which users are being promised that there will never be more than two levels between them and the most appropriate technical and service support, whether inside NCR or at a third party: all part of NCR’s overall plan to move from being a product supplier to a systems supplier and integrator, says Neches. – William Fellows