IBM’s announcement of Enterprise Systems Architecture and the MVS/ESA releases of MVS/SP and MVS/DFP are uncharacteristic in several respects. Perhaps the most striking is the short lead times – the whole lot starts arriving in August, although needless to say, the new 4381 E models, which are intended specifically to enable users to get to […]
IBM’s announcement of Enterprise Systems Architecture and the MVS/ESA releases of MVS/SP and MVS/DFP are uncharacteristic in several respects. Perhaps the most striking is the short lead times – the whole lot starts arriving in August, although needless to say, the new 4381 E models, which are intended specifically to enable users to get to grips with what may well turn out to be a radically different way of operating their systems, won’t be available until November in the US, here early next year. The other feature of the announcement that is striking is that – much more than XA – users are going to want to upgrade to the new facilities without spending too long considering first. One reason is that there are performance improvements to be had at no pre-emptive cost simply by moving to the new release of MVS, and few can resist something for nothing. But beware, if you are running a batch-only workload on a uniprocessor, you could find that the new release, far from buying you any performance improvement, actually slows you down. Another reason people will want to move is that the new software runs only on the E models of the 3090 – and if you’re machine is not upgraded to an E, its residual value will plummet: IBM expects all 3090s to have been upgraded to Es by year-end, and what Big Mother says goes in cases like that. The third reason for moving is the usual IBM big stick: MVS/ESA will be the basis for all future Systems Applications Architecture developments, and if you believe in SAA the way IBM wants you to believe in SAA, you wouldn’t want to be left out in the cold, now would you? On the face of it, the new Data Facility Product may well prove the biggest attraction of all, taking some of the hardship out of deciding where data should go on disk. The downside is that to get full advantage of the new facilities will require the usual program-by-program updating of all the user’s applications, and while the modifications are in theory easy, even easy recompilations take time, and as the procedure progresses, more and more applications will be found that really ought to be radically recast for the new architecture. And users could be forgiven for complaining that IBM could have introduced all this way back in 1981 when it introduced Extended Architecture in the first place. Why didn’t it? Unfortunately, while IBM is superbly good at a lot of things, forecasting more than two or three years ahead is not one of them. But the other piece of good news for the user is that IBM needs people to convert to Enterprise Architecture more than people need to convert, and will have to be unusually generous to ensure that the move is really attractive.
MVS/ESA: the new MVS/SP-3 The first point to understand about IBM’s new Enterprise Systems Architecture is that it is yet another kluge: IBM has not extended the address range from 2Gb to 16Tb – it has simply created a mapping facility that enables vastly increased amounts of data to be supported by applications. The expanded workspace created by IBM’s new MVS/SP Version 3 – MVS/ESA – above the 2Gb line, is only for data. The operating system enables the creation of data spaces and access to a data space is controlled by hardware: this is claimed to enhance data isolation as well as extending addressability. Within a data space, all addresses are contiguous and available to the application because the virtual storage is not divided into a system and private area, as in an address space. Application programs can use data spaces directly, and data in virtual processing is extended to allow users to map data in virtual objects into a data space, as well as into an address space. Each address space can contain up to 2Gb of combined code and data and each data space can contain up to 2Gb of data.
Hiperspaces In addition, the new operating system release supports high performance spaces or hiperspaces, designed exclusively for reading and writing blocks of data in 4Kb multiples. Data may be read from a hiperspace to an address space or written from an address space to a hip
erspace. Through efficient use of the storage hierarchy, use of hiperspaces can provide better response time for user requests – implying that they function as a sort of cache within expanded memory. Hiperspaces never reside in central storage and they come in two types. The first can be be viewed as a high-speed storage device and can be accessed only by authorised programs – and while data in a data space can reside anywhere in processor storage or on auxiliary storage and is subject to normal storage contention and paging activity, the data in this type of hiperspace will reside only in expanded storage. The second type of hiperspace, which will reside in expanded storage, if available, and can be backed by auxiliary storage, is available to all applications and can be referenced from high-level languages through the new data window facility. Data windowing services allows high-level language applications to access and scroll through large permanent data objects, real or virtual, and/or large temporary data objects. A window is a user-defined area (in the application) that holds parts of the data object. On the subject of performance of MVS/ESA, IBM says that measurements and analysis indicate the performance of MVS/SP Version 3 compared with MVS/SP 2 on ESA/370 supported hardware will be in the range of -3% to +12%. The low end of the range is in the dedicated commercial batch environment on uniprocessors, while the greatest improvement is in the IMS environment running on large multiprocessors. In addition, system response time improvements through input-output reduction can be obtained by use of data spaces, hiperspaces, and other MVS/SP Version 3 system services. Availability is planned for 1988, with staged delivery of function beginning in August. It requires an IBM 3090 Enhanced Model at the appropriate EC level that supports the IBM Enterprise Systems Architecture/370 or an IBM 4381 Model Group 91E or 92E processor. It continues to support input-output devices supported on previous MVS/XA systems – with the exception of the IBM System/3 or IBM 1130 operating as a JES2 remote workstation! It requires a minimum 8Mb of real storage, and the price is as for MVS/SP 2 releases.