Modeling Distributed Transaction Response Times As Impacted By In-Storage Buffer Accesses

The IBM S/390 Parallel Transaction Server (PTS) has introduced into the mainframe environment many aspects of distributed client/server processing. This paper begins the investigation of one of those aspects; the impact of in-storage buffer misses as additional processors are added to the complex. Additional processors are seen as a solution to both the capacity and availability problems of large transaction workloads. However, as the transactions are dynamically distributed across larger numbers of independent processors, the probability increases that the needed data will not be in an in-storage buffer, thereby increasing the transaction response time due to additional disk accesses. The concepts and principles presented can be applied to any applications moving from a single shared memory computer to multiple systems that do not share memory. This paper outlines the design of a simulation model for application designers to investigate the impacts and trade-offs of utilizing larger numbers of processors.