Abstract

In this report, we study the problem of optimizing the throughput of applications for heterogeneous platforms subject to failures. The considered applications are composed of a sequence of consecutive tasks linked as a linear graph (pipeline), with a type associated to each task. [...]