Pareto based run-time manager for overlapped resource sharing
Narasinga Rao Miniskar, Satyakiran Munaga, Roel Wuyts, Francky Catthoor · 2009
Mapping applications on a heterogenous platform in energy efficient way is a challenging problem. The goal is to find at run-time, the best assignment and schedule which minimizes the energy consumed to execute the given set of (sub-)tasks, while satisfying all given constraints. This notoriously complex problem requires a mixed design-time/run-time approach: a low and scalable overhead run-time phase assisted by an extensive design-time preparation. However, the existing approaches are limited to either two dimensional pareto handling approaches or sequential based sharing of resources or homogeneous multiprocessors. In this paper we present, for the first time, a low-overhead and multi-dimensional pareto based run-time mapping algorithm which can share the platform resources in overlapped basis.