At the UK Computer Measurement Group Bill Mullen of BGS Systems Inc spoke to delegates about the delights the MVS/ESA operating system architecture has opened up for system designers and users. Expanded Storage technology on the 3090 processor complex has eased many performance problems, for example reducing the time required to transfer pages to and […]
At the UK Computer Measurement Group Bill Mullen of BGS Systems Inc spoke to delegates about the delights the MVS/ESA operating system architecture has opened up for system designers and users. Expanded Storage technology on the 3090 processor complex has eased many performance problems, for example reducing the time required to transfer pages to and from Central Storage from milliseconds to microseconds, and offering a faster alternative to use of auxiliary storage while freeing that precious commodity Central Storage. On the other hand, as soon as additional function is provided on an installation, operating system subsystems and application system Address Spaces disobligingly grow to fill it, creating further Virtual Storage Constraint Relief problems, while the increase of transaction volumes and complexity has added to I/O drag. Dataspaces The Data-in-Virtual facility was introduced with the MVS/XA environment to try and address the processor storage data problem, and Mullen argued that those Virtual Storage Access Method linear data sets offer a basis for the future elimination of input-output and performance improvements. Meanwhile, three new facilities are available under the ESA/370 architecture: Dataspaces, Hiperspaces, and Large Virtual Buffering. Mullen discussed each of these in turn and began by describing dataspaces as an MVS Address Space that didn’t quite make it, since both are created in much the same way but the former do not have system areas and therefore provide the user with 2Gb of addressable virtual storage. The user has two constraints imposed upon the number of dataspaces that can be created and used. Firstly, only 8,192 dataspaces can be active at any one time; secondly, only 256 concurrent dataspaces can be owned by an address space, of which three are reserved for the system, leaving the user with a potential 253 unique dataspaces. Dataspaces can reside in virtual storage (defined in 4Kb byte addressable blocks) with real storage backing, on Expanded Storage pages or on auxiliary storage. Data manipulation is less complicated for data spaces and MVS cross-memory access in MVS/ESA, as it has 16 hardware registers, called Access Registers which correspond to the 16 General Purpose Registers. Other enhanced features in MVS/ESA include a Library Lookaside facility implementation with allowances for directory management of any type of partitioned data set, while the Virtual Library Facility operates as an independent address space for storing named data objects in the VLF address space as well as a server for storage and retrieval services to subsystem dataspaces. Mullen claimed that because the size of data object storage and its effects upon real storage frame availability and system paging rates is the main worry with most installations, the Virtual Library Facility offers immediate scope for input-output elimination for data sets. As for the future uses of dataspaces, Mullen believed that they could be used for MVS control block and data storage that is normally available in the Common System Area, thereby cutting the virtual storage requirements to below the 16Mb line and offering potentially larger user private areas. Secondly, Mullen suggested that a Job Entry Subsystem dataspace could be used for spooling purposes, so that small print data sets, from the TSO environment for example, could be directed to such spool dataspaces and subsequently retrieved using, say, the ISPF 3.8 screen option. Lastly, he recommended that dataspaces might be used as a temporary storage area for SMF records, as most people do not need SMF data again two minutes after it is written. The second main facility offered by ESA is the Hiperspace which is a dataspace derivative. Hiperspace pages live either in Expanded Storage or in auxiliary storage but never in central storage and are of two types: cache or scrollable. Data in the cache type is not guaranteed and must have a permanent disk source. According to Mullen it is a common fallacy that pages in cache-type Hiperspaces are non-migratable, but if Expand
ed Storage frames are depleted to a critical state such pages will be migrated. Scrollable Hiperspaces can be created by any task, reside in Expanded Storage and can be migrated. Mullen has found that standard Virtual Storage Access Method data sets can be mapped directly to Hiperspaces through the Local Shared Resource buffer pool facility if the DFP Release 3.1 product is used for authorised tasks. A further capability that Mullen has discovered is that the Build Virtual Resource Pool macro can get the token of Hiperspace in which the Local Shared Resource Pool will be constructed. A capability that will prove to be of value to users of CICS/VS 2.1 and IMS/VS 2.2. The third main facility offered by ESA is data windowing which provides callable subroutine interfaces for high level languages such as Cobol and Fortran, thus giving access to temporary or permanent data objects in both dataspaces and hiperspaces. This facility deals with Virtual Storage Access Method linear data objects and is built around the initial Data-in-virtual facilities available in MVS/XA; consequently, for the Expanded Storage to be used to its full capability, Mullen believes that the designer and programmer must have a thorough grasp of the Data-in-Virtual processing concept. Difficult to install Mullen’s overall conclusion was that ESA does provide opportunities for enhancing the performance of large systems, so long as users: were conservative in their input-output reduction expectations since sequential type access methods cannot be suppor-ted in dataspace and Hiperspace; remembered that DB2 is a memory hungry application which may push them over the edge of Expanded Storage capabilities; and were careful with their usage of Library Lookaside and Virtual Library Facility because potential decreases in input-output may be more than offset by the increase in paging for the system. Mullen believes that the MVS/ESA operating system architecture has been slow to prove itself because it is difficult to install, get up and running, and then use. However, the more it is used and explained by people like Mullen, the more popular it is likely to become.