Formal Approach for GPU Architecture Schedulability

Analyse data issued from Social networks, large scale wireless networks, .. is computation intensive, and submitted to soft or hard real time constraints. The main characteristic of these kind of applications is that the execution time is greater than the deadline. Thus, mono processor architectures can not satisfy real time requirement of this type of applications. Multiprocessors nowadays architectures consists of numerous processors on one chip and allows to run tasks in parallel manner and can handle the overrun of these applications. In this paper, we present a novel online scheduler for real times tasks where execution time is greater than deadline. As application example, we use MapReduce Real time environments to extract simulation parameters and run tests on simS simulateur. Models & Optimisation and Mathematical Analysis Journal Vol.02 Issue 02 (2014) 3 Formal Approach for GPU Architecture Schedulability ZOUANEB Imane, BELARBI Mostefa University of Ibn Khaldoun Computer Sciences Department Tiaret, Algeria LIM Research Laboratory i_zouaneb@yahoo.fr, belarbimostefa@yahoo.fr CHOUARFIA Abdellah University of Sciences and Technology Oran Mohamed Boudiaf Computer Sciences Department Oran, Algeria LIM Research Laboratory chouarfia@univ-usto.dz Abstract— Parallel application modelling and specifying is not an easy task to do because it treats tasks scheduling and time evolution. Graphics processing Unit is one of the main architectures that guaranties parallel execution. Event B is a skilled formal language based on sets theories. Our goal is to model and to specify the parallel execution of programs on GPU using Event B & RODIN platform. We are interesting to timing and scheduling of tasks on GPU. Key-Words : Parallel application, GPU, Formal specification, Timing, Scheduling, Event B. Parallel application modelling and specifying is not an easy task to do because it treats tasks scheduling and time evolution. Graphics processing Unit is one of the main architectures that guaranties parallel execution. Event B is a skilled formal language based on sets theories. Our goal is to model and to specify the parallel execution of programs on GPU using Event B & RODIN platform. We are interesting to timing and scheduling of tasks on GPU. Key-Words : Parallel application, GPU, Formal specification, Timing, Scheduling, Event B. Models & Optimisation and Mathematical Analysis Journal Vol.02 Issue 02 (2014) 4 Based refinement Verification platform for QNoC Architectures