Mapping dynamic applications on MPSoC systems
Zubair Wadood Bhatti, Davy Preuveneers, Roel Wuyts, Yolande Berbers · Lirias · 2010
My PhD research will investigate the optimization of executing dynamic applications on heterogeneous multi-processor systems-on-chips (MPSoCs). The optimization is aimed at minimizing the resource requirements through efficient deployment of software onto hardware. The deployment will be carried out by run-time environment assisted by input from a pre-deployment exploration tool chain. The targeted application’s resource requirements cannot be predicted beforehand due to dependences on user input and application data. On one hand a pure design time approach to the allocation and scheduling parts of applications to different cores and memory hierarchies will have to stick to the worst case estimates and will therefore result in a suboptimal resource usage. On the other hand a pure run-time approaches either has a very high overhead penalty [1] or do resource management at a very abstract level with a high granularity and a sub-optimal performance. In our research we assume that the application is represented as an annotated distributed model of computation such as the grey-box model [2]. The grey-box model can be described as a fine grained Khan Process Network (KPN) annotated with information such as the time required and energy consumed for running a process on a specific processing element. The