Rechercher des projets européens

From Inherent Concurrency to Massive Parallelism through Type-based Optimizations (UpScale)
Date du début: 1 févr. 2014, Date de fin: 31 janv. 2017 PROJET  TERMINÉ 

In a radical paradigm shift, manufacturers are now moving from multicore chips to so-called manycore chips with up to a million independent processors on the same silicon real estate. However, software cannot benefit from the revolutionary potential power increase, unless the design and code is polluted by an unprecedented amount of low-level, fine-grained concurrency detail.Concurrency in mainstream object-oriented languages is based on multithreading. Due to the complexity of balancing work evenly across cores, the thread model is of little benefit for efficient processor use or horizontal scalability. Problems are exacerbated in languages with shared mutable state and a stable notion of identity -- the very foundations of object-orientation. The advent of manycore chips threatens to make not only the object-oriented model obsolete, but also the accumulated know-how of a generation of programmers.Our vision is to provide the means for industry to efficiently develop applications that seamlessly scale to the available parallelism of manycore chips without abandoning the object-oriented paradigm and the associated software engineering methodologies.We will realise this vision by a breakthrough in how parallelism and concurrency are integrated into programming languages, substantiated by a complete inversion of the current canonical language design: constructs facilitating concurrent computation will be default while constructs facilitating synchronised and sequential computation will be explicitly expressed. UpScale will exploit this inversion for a novel agile development methodology based on incremental type-based program annotations specifying ever-richer deployment-related information, and for innovative type-based deployment optimisations both at compile time and at runtime in the runtime system devised in UpScale for massively parallel execution.The targeted breakthrough will profoundly impact software development for the manycore chips of the future.



Projet Website

3 Participants partenaires