Erik Cota-Robles

Schedulability analysis for desktop multimedia applications: simple ways to handle general-purpose operating systems and open environments

Unlike traditional real-time applications, multimedia applications are commonly deployed on top of general-purpose operating systems in open environments. Because Rate Monotonic Analysis (RMA) assumes that the total system workload is known in advance and uses worst case values for operating system overhead (e.g., context switch time), it has not been usable for personal computer and workstation based multimedia applications. We extend RMA to model systems where these OS services have highly non-deterministic timing behavior and where the total system workload is not specified in advance. We validate our techniques by modeling a video telephony application based on Intels ProShare(R) technology.

The design and implementation of the intel/spl reg/ real-time performance analyzer

Modern PCs support growing numbers of concurrently active independently authored real-time software applications and device drivers. The non realtime nature of PC OSes (Linux, Microsoft Windows, etc.) means that robust real-time software must cope with hold-offs without degradation in user perceivable application quality of service. The open nature of the PC platform necessitates measuring OS interrupt and thread latencies under concurrent load in order to determine with how much hold-off the application must cope. The Intel/spl reg/ Real-Time Performance Analyzer is a toolkit for PCs running Microsoft Windows. The toolkit statistically characterizes thread and interrupt latencies plus Windows Deferred Procedure Call (DPC) and kernel Work Item latencies. The toolkit also has facilities for analyzing the causes of long latencies. These latencies can then be incorporated as additional blocking times in a real-time schedulability analysis. An isochronous workload tool is included to model thread and DPC based computation and detect missed deadlines.