A framework for design-time configuration of a runtime management system for many-core architectures

The Many-Core Architectures are the next big turning-point in the development of architectures. We have to be ready to make the most of them by exploiting the advantages that they can offer, but it is not trivial to efficiently utilize the available computing power. The currently used solutions, by themselves, do not scale well with great number of cores, particularly in the presence of concurrent parallel applications. The goal of this thesis is the development of an efficient Runtime Management System that allows to exploit the advantages offered by the Many-Core Architectures through the dynamic choice of the parallelization level of the applications. In order to be used with a wide range of applicative scenarios, it has also been implemented a Design-Time Configuration Framework that helps finding the best configuration of the Runtime Manager parameters. Experimental results have been conducted on a real Many-Core system, to analyze the difference in system throughput and individual application performance when using the Runtime Management System. They show that it is introduced low overhead and that it is obtained a consistent performance improvement in the execution of multiple concurrent running applications.