Home > Research > Étudiants/Post-docs > Amaury Graillat (M2R)

Amaury Graillat (M2R)

Towards Semantics-Preserving Implementation of Synchronous Programs on Many-Core Architectures

Thursday 1 December 2016

Years: 2015 (M2R )
Co-supervisor: Florence Maraninchi
Subject: Towards Semantics-Preserving Implementation of Synchronous Programs on Many-Core Architectures

Reactive systems are systems that constantly react to their environment. One must guarantee that the computation time of the system is bounded and matches the input-output latency imposed by the environment (for instance the period of measure of the altitude for a flight control system). We talk about multi-periodic systems when the nodes of a program are running at different periods. Today, reactive programs are more and more complex, and we reach the limit of the single-processors in term of computation. Consequently, the size of the program is limited to match the time requirements. Most of the processors are multi or many-core. But they are complex and the Worst-Case Execution Time of a program executed on these architectures does not guarantee the time requirement. Yet, the Kalray MPPA many-core architecture is predictable and allows to bound both the execution and the transmission time between the many-core clusters.

Our purpose is to run synchronous programs on many-core architectures with bounded delays. On one hand, we consider a synchronous program composed of several nodes communicating together, on the other hand a many-core chip composed of several clusters and a Network-On-Chip (NoC). The first problem is to find a way to map the nodes on the clusters. The other problem is to give the semantics of the communication between the nodes. The semantics must be deterministic and preservable when implemented on the many-core architecture to guarantee equivalence between the synchronous program and the implementation. We give a deterministic semantics to communications between the nodes of different periods. We give the implementation of a multi-periodic program on the Kalray MPPA. Finally, we evaluate our solution by testing. Our method provides a key element toward the deterministic implementation of reactive programs on many-core architecture: a way to preserve a centralized deterministic semantics of a multi-periodic program, when the program parts exchange information via the NoC.

Attached documents

Valid XHTML 1.0 Transitional
SPIP | | Site Map | Follow site activity RSS 2.0
Graphic design (c) styleshout under License Creative Commons Attribution 2.5 License