The IBM viewpoint on the benefits of Expanded Store, Enterprise Systems Architecture In last week’s report on IBM UK’s Enterprise Systems Forum, we highlighted IBM the company’s responses to some of the burning questions that are exercising the vast army of people that co exist with IBM, be it as users, business partners or competitors (CI No 1,166). But when IBM calls such a meeting, only the foolish dismiss what the company has to say on its own account as mere flackery and puff for its own products. Those with a need to know at all times where IBM is at and where it intends to go press the attendees with questions about what was highlighted being fully aware that what IBM chooses not to talk about can be as important as what it highlights. One key part of the formal presentation appeared in last week’s report – IBM’s anxiety to stress the benefits of multiprocessing when it is done the IBM way. We thought that all the arguments about whether adding a third or a fourth processor to a complex was worth the money had been laid to rest with the 308X family. There were problems with the 3084Q in the early days, but by the time the 3090s started shipping in volume, there were plenty of 3084Q users multiprocessing away in single image very happily, and IBM has done a fair bit more since to make multiprocessing more cost effective – and remember that partitioned machines, such as a 3090-400S run as two 200s, are particularly attractive because the user has the functionality of two completely separate machines while having to pay only one set of software licences. IBM seemed to be saying last week that because of the way the machines were configured for multiprocessing, a 200S actually runs more efficiently than a 180S. What IBM of course didn’t repeat was the dictum of that wise and charming old owl, the developer of the 360, Dr Gene Amdahl, that the best place to start is always with the most powerful uniprocessor you can produce. Expanded Storage means reliability With characteristic bullishness, Enterprise Systems Marketing man Nigel Seymour-Dale kicked off the Enterprise Systems Forum by claiming that current growth opportunities in the UK would see IBM through well into the next century. The twin challenges, he argued, were producing the raw horsepower, and handling the data, offering MVS/ESA and DFSMS as IBM’s respective answers. DFSMS is of course the Data Facility Storage Management Subsystem that takes the organisation of data on disk out of the hands of the user and takes it upon itself to decide the physical layout in a way that should prove more efficient than any individual could arrange it. On specific hardware developments, he cited multiprocessors (CI No 1,666) and storage developments, notably Expanded Storage. Here, the clear message was that reliability, not performance should be regarded as the main issue, as Expanded Storage is designed to meet a different requirement from Central Storage, in a different way. According to Seymour-Dale, the role of Central Storage is to meet high user requirements. By contrast, IBM has designed Expanded Storage to meet average user demand, and, via multiple-bit checking, has prioritised reliability over performance. As regards competitors, Seymour Dale acknowledged that Hitachi Ltd had announced plans to split its Central Storage, while Fujitsu Ltd had indicated that it would ship Expanded Storage on the Amdahl Corp 5990 series marketed in Japan by Amdahl’s biggest shareholder and technology partner. Adressing the latter, however, he argued that IBM’s technical approach to the issue had distinct advantages. Under plans outlined by Fujitsu, Expanded Storage on the 5990 range will be added to the system control element. Seymour-Dale claims that this slows down the processor, because each request for data has to go through the system control element before it can reach Expanded Storage. IBM’s alternative is to attach Expanded Storage to Central Storage, so that requests by-pass the system control element. ESA means much better price-performance Turning to Extended
Systems Architecture, Seymour-Dale claimed that ESA has nothing to do with removing constraints. The role of memory-based computing, he insisted, is to provide users with a range of options to gain a more productive system, leading to potential price-performance boosts of up to 40%. Key advantages claimed are, firstly, that ESA reads from Expanded Storage, pushing up system response time. Second, four-way throughput, and the fast write and dual copy advanced functions on the 3990-3 disk controller, combine to make the system more productive, without performance degradation. Finally, ESA’s potential to reach an input-output rate of 200, means that it is an appropriate architecture for business transactions. A number of performance tests from IBM’s Internal Systems Laboratory were held up as evidence. A comparison of a systems running MVS/ESA and MVS/XA showed a reduction in response-time to 0.84 seconds from 2.23 seconds, and a productivity rise of 62% – the key to solving the application program bottleneck, claimed Senior Consultant Noel Bradbear. Additional test claims included a 35% improvement in processor usage, and a 7% gain in response time with Hiperspace/IMS, and a 28% and 8% gain in processor use and reponse time respectively with CICS/VSAM running under ESA, using 420Mb of Expanded Storage. Finally, the use of the DFSort function on a 3090-300E showed a 17% boost in CPU utilisation, and a gain in the input-output rate of around 86%. SAA is here, now and available Despite the fact that, to date, software developers attempts to develop SAA-looking applications appear to have focussed on CUA Common User Access-compliance, IBM says that the information for developing SAA-compliant applications is there. In particular, it points to the availability of specifications for both the Common Programming Interface and the Common Communications Interface components. However, it adds that like all living architectures, SAA will continue to develop.
