Thread-level parallelism and interactive performance of desktop applications

Multiprocessing is already prevalent in servers where multiple clients pres ent an obvious source of thread-level parallelism. However the case for mul tiprocessing is less clear for desktop applications. Nevertheless, architec ts are designing processors that count on the availability of thread-level parallelism. Unlike server workloads, the primary requirement of interactiv e applications is to respond to user events under human perception bounds r ather than to maximize end-to-end throughput. In this paper we report on th e thread-level parallelism and interactive response time of a variety of de sktop applications. By tracking the communication between tasks, we can foc us our measurements on the portions of the benchmark's execution that have the greatest impact on the user. We find that running our benchmarks on a d ual-processor machine improves response time of mouse-click events by as mu ch as 36% and 22% on average-out of a maximum possible 50%. The benefits of multiprocessing are even more apparent when background tasks are considere d. In our experiments, running a simple MP3 playback program in the backgro und increases response time by 14% on a uniprocessor while it only increase s the response time on a dual processor by 4%. When response times are fast enough for further improvements to be imperceptible, the increased idle ti me after interactive episodes could be exploited to build systems that are more power efficient.