Influence of Channel Loading on Computer System Performance

Abstract

Computers have limited capabilities to handle simultaneous data traffic from many I/O devices. Especially for unbuffered I/O devices this may lead to loss of data, error interrupts, and costly error recovery procedures.

This paper uses some basic concepts from the state-of-the-art worst case analysis method to arrive at probability densities for individual channel loads. Total load sums are computed by combining individual loads using multiple, iterative convolutions. Final results of the analysis are probabilities of overruns and mean time between overruns.

In addition, an interactively operating channel configuration aid is described which incorporates the novel analysis methods and generates values for I/O interference, worst case load sums and estimates of mean time between overruns.