Data-flow timing models of dynamic multimedia applications for multiprocessor systems

Efficient implementation of complex state-of-the-art multimedia applications on a system-on-chip requires a multi-processor architecture. For this purpose, we have developed modeling techniques for application mapping on predictable multiprocessor systems. In the mapping experiments, it was found that the static nature of synchronous data-flow graphs is not suited for multimedia systems, which are increasingly characterized by dynamic workload. Therefore, we present an extension that also captures the dynamic properties of multimedia applications. We chose MPEG-4 shape-texture decoding, as a complex example for the modeling of an application with dynamic behavior. Experiments with a set of five test sequences show that our model has a high accuracy (3.4%). A key advantage of our modeling method is that it allows to analytically derive the performance characteristics.