NCR V-8500m computers offer a simple, flexible way to improve performance by upgrading to multiprocessing whenever applications outgrow the constraints of a single- processor environment. Typically, a balanced dual-processor system provides from 1.5 to 1 .8 times the performance of a comparable single-processor system, depending upon the nature of the job mix. Although dual processing is the principal operating mode, system architecture and operating software can accommodate up to four processors in one system.
NCR multiprocessing systems are tightly coupled, share all resources equally, and operate under a single copy of the VRX-MP operating system. This means that existing V-8500 programs and files can be transferred directly from single- to multiprocessor operation without having to divide or restructure tasks into separate workloads. VRX-MP automatically dispatches units of work from a common pool to available processors, continually leveling the load among processors for maximum efficiency, without any operator intervention.
In addition to the improved performance derived from dynamically balanced, shared-load processing, multiprocessor systems also offer greater reliability through inherent redundancy. Should a non-critical element fail, operation will continue in a degraded mode, usually without any operator intervention being required.
Individual elements can also be 'removed" for maintenance without shutting down the entire system, and diagnostics can be performed through a second console concurrent with normal operations. If necessary, the console used for diagnostics can be remotely located at a local, regional, or national NCR service center and connected to the system over a common carrier link.
Asymmetrical Multiprocessing Systems
Unlike other multiprocessing systems, V-8500MP systems do not have to be symmetrical. That is, different processor models, with different size memories, can be connected into a single system if a user's requirements dictate.
Low-cost Peripheral Back-up
Peripherals, like memory, are considered as part of a common resource pool, so a single device can back up any other device of its type in the system, resulting in less cost to ensure peripheral redundancy.
NCR multiprocessing systems can be reconfigured whenever necessary for multiple uniprocessor operation by simply reloading firmware and making any peripheral reassignments that may be required. This flexibility permits users to tailor environments for changing conditions as they arise.
NCR V-8500MP architecture is a logical extension of NCR's 8500 family bus architecture. Bus architecture groups independent subsystems into an integrated base system and allows expansion of the base by incorporating additional subsystems. Multiprocessing systems extend this architecture by incorporating high-speed Interbus Communication Adapters (ICA) that allow operating system software to view the total system as a pool of resource elements available for load-leveling. The fact that multiple processors are in the system is virtually transparent to the operating system and totally transparent to user applications.
VRX-MP is the multiprocessing extension of powerful, flexible NCR Virtual Resource Executive software. It schedules and runs up to 35 jobs concurrently, automatically allocating memory, processor time, and I/O as needed. Each job can consist of one or more tasks, or programs, and the user can exercise as much or as little control over job processing as he wishes through priority assignments.
VRX-MP has been designed to relieve the operator of much of the scheduling burden and to simplify the task of operating a multiprocessing system by allowing him to treat it basically as if it were a uniprocessor system. Systems employ a single job input stream and can be run through a single operator control console, an approach that is easier and far less error-prone than controlling a large, complex system from multiple consoles. System messages are presented in a straightforward 'menu" format. Menus allow the operator to page quickly and easily through a hierarchical structure leading from general to specific information about the system or any of its individual elements.
Subsystems can execute at the same time in different segmented tasks of the same job or different jobs. Operating system software keeps all processors equally busy by spreading the workload across the resources available based solely upon the priorities of various active tasks. Any processor in the system can execute any task as it is dispatched, to achieve maximum performance through continuous load-leveling.
VRX-MP permits large applications to be broken down into sets of small, logical units. Each unit can be designed, coded, and implemented completely independent of every other unit, and the addition or deletion of a unit will have no effect on any other unit. Each unit has its own memory and tile protection, is individually recoverable, and can be initiated, run, and terminated dynamically. Units can communicate with each other when necessary through a separate communication facility.
In a transaction processing environment, VRX permits individual VRX COBOL application units to be replicated and then executed asynchronously, employing a parallel multi-tasking feature. This facility allows each unit to use available resources most effectively, ensuring optimum processor utilization under all conditions, and especially during peak transaction processing periods. Replicas will execute completely independent of one another, and application programmers can write programs as if they were to perform a single task. If data is to be shared among replicas, the only additional programming effort required is the inclusion of a few additional statements and data declarations.
VRX-MP Error Recovery
VRX-MP provides multiple levels of error recovery for multiprocessing environments, particularly transaction processing environments, which must be generally insensitive to errors. Errors will not cause program termination until all applicable levels of recovery have failed to restore normal operation. Insofar as possible, the effects of an error will be confined to a single transaction. Recovery facilities, which encompass hardware- and software-detected user errors as well as system hardware, software, and firmware errors, have been designed to ensure maximum availability, prevent undetected data loss, and allow applications to resume operation without manual intervention.