It looks as if September’s CMP-Gartner Group Inc-sponsored Commercial Parallel Processing Conference is going to be packed with new high-end stuff. But maybe they should be calling it the high-end symmetric multiprocessing, symmetric multiprocessing clustering, parallel processing, commercial massively parallel processing and supercomputer conference, judging by the folk that graced last week’s warm-up event in […]
It looks as if September’s CMP-Gartner Group Inc-sponsored Commercial Parallel Processing Conference is going to be packed with new high-end stuff. But maybe they should be calling it the high-end symmetric multiprocessing, symmetric multiprocessing clustering, parallel processing, commercial massively parallel processing and supercomputer conference, judging by the folk that graced last week’s warm-up event in New York City. General consensus is that after decision support, warehousing and data mining, high-end and highly scalable technology will find its way into first messaging and then intelligent transaction solutions. By the year 2000, parallel technology will be endemic, Gartner claimed. Hewlett-Packard Co has a new high-end strategy it’s calling Enterprise Parallel Clusters; although there’s not too much parallelism apparent in the traditional sense. Eschewing a shared-nothing approach the company will, from September, begin tying its commercial symmetric multiprocessing HP-UX servers together for high-end customer requirements, using high-speed Fibre Channel technology currently deployed on its workstations.
By the way, its re-badged Convex Computer Corp Exemplar scalable parallel processing systems are strictly technical engines, it said ( CI No 2,554). The clusters will also employ a a programming model the company likens to auto-focusing camera techniques, where the lens will automatically refocus to accommodate a moving target. It says that its model will predict the kinds of questions users want to ask and formulate data in readiness. Like Hewlett-Packard, Digital Equipment Corp does not have a shared-nothing parallel system in its sights, planning to build on the raw speed of its 64-bit Alpha AXP for high-end requirements. The company said it has got as many as 200 people now engaged on decision support solutions and – to the horror of the Transaction Processing Council – claims its 12-way AlphaServer 8400 TurboLaser will record 9,000 tpmC by September. At that time it will unveil its 100Mbps Memory Channel interconnect it says will link symmetric multiprocessing nodes with a latency of less than a millisecond over bog-standard SCSI. Memory Channel will begin aboard a multichip set before it gets down to a single chip implementation. DEC said it went a long way down the Non-Universal Memory Architecture model being used by the likes of Data General Corp before it realised the kind of restraints this approach imposes on developers and just how well latency could be reduced with its home-grown technology. Both DEC and Hewlett-Packard will build on their existing Unixes to provide a single system image of each node or cluster, not the system as a whole. DEC said it is currently working on a deal that could result in a 1 TFLOPS, $150m system sale. Meantime, AT&T Global Information Solutions is preparing the long-delayed system implementation of its next-generation BYnet interconnect for announcement around the same time-frame. The project, still known internally as 3700 – though it won’t come to market as that – is a massively parallel processing hardware and software development running the Teradata database with high-speed fibre links between nodes. AT&T said it is not taking Intel Corp’s SPP parallel system OEM (CI No 2,655), claiming its own unit could interrogate up to a Petabyte, or 1,000Tb of data. It still claims the world’s largest commercial database, the 4Tb WalMart Inc stores system that is moving up to 7Tb.