Profiling Driven Scenario Detection and Prediction for Multimedia
Applications
Modern multimedia applications usually have real-time constraints and they are implemented using heterogeneous
multiprocessor systems-on-chip. Dimensioning a system requires
accurate estimations of resources needed by the applications.
Overestimation leads to over-dimensioning. For a good resource
estimation, all the cases in which an application can run must
be considered. To avoid an explosion in the number of different
cases, those that are similar with respect to required resources
are combined into, so called, scenarios. This paper presents
a method and a tool that can automatically detect the most
important variables from an application and use them to define
and dynamically predict scenarios, with respect to the necessary
time budget, for soft real-time multimedia applications. The tool
was tested for two multimedia applications. Using a proactive
scenario-based scheduler based on the scenarios and the runtime
predictor generated by our tool, the cycle budget over-estimation
decreases with up to 83.50%, paying an acceptable cost of up to
1.74% in the number of missed deadlines.
|
