Measurement and evaluation of multimedia I/O performance

I/O is becoming the bottleneck in computer systems as the disparity between the speed of the I/O system and the CPU continues to grow. This research addresses the performance evaluation of I/O, with special emphasis on multimedia workstation-class systems. First, multimedia I/O activity is shown to be bursty with correlation across multiple device types (such as video, network, and disk). Second, several novel instrumentation and measurement techniques are developed to allow accurate measurement and analysis of multimedia I/O activity. Third, an I/O resource contention model is created that incorporates measurements of I/O device interaction. Fourth, the I/O signature is developed as a technique for evaluating and predicting application I/O performance. Fifth, a multimedia benchmark and multimedia benchmarking tool are constructed to allow quantitative comparison of I/O performance on different machines at several levels of detail. Sixth, the techniques for instrumentation, measurement, and analysis are incorporated into a single tool called Priority, providing automatic, flexible evaluation of I/O, with unique capabilities for evaluating multimedia systems.U of I OnlyETDs are only available to UIUC Users without author permissio