Sheffield Computer Group Plc, which has changed its name from Sheffield Micro, is expanding with the acquisition of two ICL Traders, Atlantic Data Systems Ltd of Frome, Somerset specialising in ICL-based office systems, and Information Engineering Services Ltd of Manchester, doing manufacturing management systems. Terms were not disclosed. It has also become a founder member […]
Sheffield Computer Group Plc, which has changed its name from Sheffield Micro, is expanding with the acquisition of two ICL Traders, Atlantic Data Systems Ltd of Frome, Somerset specialising in ICL-based office systems, and Information Engineering Services Ltd of Manchester, doing manufacturing management systems. Terms were not disclosed. It has also become a founder member of ICL’s Computers In Manufacturing Partnership, which offers medium-sized manufacturers expertise in ICL Unix manufacturing systems. IBM Enterprise Systems Forum Correcting all those misconceptions about Enterprise Systems Architecture For want of anything really new to discuss, IBM UK this week staged an Enterprise Systems Forum conducted by Nigel Seymour Dale, Enterprise Systems Marketing, and Noel Bradbear, senior consultant (Systems Engineering). Their main message was that Enterprise Systems Architecture is not a new level of performance intended for users who have hit the ceiling on the old architecture, but a new option that is available to – and intended for – everyone from the smallest 4381 user up. They also discussed specific aspects of the new IBM large systems and answered questions of moment. Software budgets will continue to rise Software budgets will continue to rise was the message from Seymour-Dale, who argued that network and data management software were vital for handling increasingly complex volumes of data. At a more practical level, he provided new insight into IBM’s restuctured software pricing, by insisting that by replacing once-and-for-all charges with an upfront charge and a licence payment that has to be renewed every year, IBM was protecting its users from being locked into any one version of a software product – clearly the practice in the micro software industry of offering new releases to existing users at a deep discount hadn’t occurred to IBM. And evidently, the new policy has nothing to do with the fact that IBM regularly puts up its prices well above the rate of inflation… 3090s are slewed to multiprocessing Exposing yet another of those design choices, Seymour-Dale admitted that uniprocessor efficiency had been sacrificed on the multiprocessing altar. The multiprocessor development strategy had been adopted to meet user requirements for more power he explained, implying that, with a current growth rate of between 15% and 20%, uniprocessors simply weren’t up to the job. The system control element on a 3090 processor occupies one arm of a crossshaped central processor complex. In performance terms, the ideal position for a uniprocessor system control element would be right in the middle of the central complex. However, IBM uses the former configuration to ensure that when two processors are joined together, forming a mirror image, the two system control units lie directly opposite each other. According to senior IBM consultant Noel Bradbear, this configuration boosts multiprocessor performance, by providing the optimum conditions for communication between the two processors. And the conclusion? When IBM multiprocessors are benchmarked the performance advantage really begins to show. How to benchmark against Amdahl 5990s Despite repeated refusals to publish its proprietary benchmarks it says that the amount of research that goes into creating a set of IBM benchmarks is so great that it has to be regarded as a significant asset of the company, so why should it give it away? – IBM shows no scruples in using them to attack its competitors, or using the results as the basis of customer presentations. However, the latest – albeit subjective – claim appears to be based on an even more closed formula, known simply as IBM’s historical view. Broadly speaking, this means extracting a performance trend, based on Amdahl’s claims for earlier machines and how those panned out in practice, and applying it as a benchmark substitute, to the latest Amdahl offering. The historical view of the Amdahl 5990-1400 is that it will lag some 27% in performance terms behind a 3090-600S. In fact, the historical view places the
5990-1400 mainframe roughly on a par with a 3090-500S. The performance lag is illustrated, via a tri-colour block chart, no doubt on its way to users. Dedicated 3090 CPUs? Great idea, but.. Still on the subject of multiprocessors, whatever happened to the idea, first mooted some five years ago, that by adding extra microcode to 370 processors you could have, say, a dedicated database or communications processor within a multiprocessor complex? IBMers talked enthusiastically about the concept at briefings in the early 1980s, but now points feebly to its Vector Facility, and says that it undoubtedly has the technology clearly, after mature throught, it just went off the idea. How much Expanded Store does ESA need? Almost everyone knows by now that you need vast quantities of expanded storage to exploit Enterprise Systems Architecture hence the recent ESA incentive in the shape of a drop in expanded storage prices. Pressed to provide a more accurate guide, IBM describes 400Mb as an average requirement, but adds that the figure varies depending on what you want and where you’re coming from. Addressing the polar extremes, it went on to suggest that 192Mb on a 3090-180 base unit, and 512Mb on a 3090-600S, would suffice. And when will the delights of ESA become available to VM users? Short-term, the answer is that VM is directed for a different market from MVS, but that you can run an inefficient form of ESA with Release 2 of VM/XA, shipping now. Long-term, the answer is not this year – which likely means 30 months after MVS/ESA. No microcode problem on the 3990-3 And when will IBM fix that microcode problem on the 3990-3 disk controller that is limiting users’ appetites for 3380K disk drives? According to Nigel Seymour-Dale, there’s no specific problem, it’s just that microcode takes a long time to write. In a piece of inspired logic, he also rebuffed any notion of a delay in delivery, by arguing that as IBM had never said when the product would be available, it could hardly be accused of having fallen behind schedule. Short shrift was also given to suggestions of paralysis in the IBM disk drive market as users wait to see the new drives IBM has on the stocks – We don’t see any paralysis. Current demand for IBM disks is very good, to be precise. So why are users, waiting for IBM to fix the 3990, and announce its new disk drive, stocking up on used or cheap NatSemi stop-gaps? Because NatSemi makes very good disks. The 9370: the 8100 reborn as a 370 There are as many roles for the 9370 these days as there are IBMers to answer the question. Despite the fact that the 9370 is now an Enterprise System/9370 and the briefing was on Enterprise Systems, the machine was outside the brief of the two-man team, so it was left to the regular IBM information machine to explain that while running CAD/CAM applications and Geophysical Facilities Information System software has proved a prime attraction, it is now aimed mainly at large 370 users who want to put small CPUs into remote locations – in other words it’s now the successor to the 8100.