Processing Times for Segmented Jobs with I/O Compute Overlap

Publisher Website
Google Scholar
Abstract
Compute-output processing times are determined for n -segment jobs that are preloaded into main storage and processed with overlap. A queueing model with tandem servers is utilized for the performance analysis. In particular, the solution presented involves determination of the transient response for a batched arrival of n segments to be processed through two stages of tandem service with unlimited output buffering. The performance results provide insight into conditions arising in systems consisting of a single CPU and I/O channel with overlap capabilities. Two cases, single-segment overlap and unlimited overlap, are considered. Segmental compute and output (or input) service times are taken to be exponentially distributed; however, the approach is not limited to the exponential case if service is independent. The ratio of mean output time to mean compute time is varied to explore the full range between compute-bound and output-bound extremes. Final results are presented as relative gain over sequential processing.