Multi-Core Paralle Heterogeneous Programming – MCOREPHP

In the past few years, all of the major processor vendors adopted multi-core architectures as the main way to improve the processor performance. Within the context of Grid, cloud and other high performance computing infrastructures that traditionally used parallel and concurrent programming, the use of multi-core architectures introduces an additional level of complexity: the intra-node parallelism. The currently available multi-core architectures rely on the well-known SMP model, but the new trends in processor development target asymmetric multi-core architectures (cores with the same instruction set, but different performance in terms of clock speeds, CPUs versus GPUs, etc). In this MCorePHP project, we would like to investigate certain methods and techniques that help simplify the parallel programming without sacrificing performance. Some of the main areas we would like to touch are scheduling, synchronization and proper use of the multi-core architecture features. This requires new programming models, in particular for managing concurrency and memory sharing within the processor architecture. But achieving good performance isn’t our only goal. Multi-core systems tend to become harder to handle as the number of cores per chip increases. Therefore, we need a safe, dependable, autonomic way of developping applications on multi-core processors, but also on multilevel infrastructures including multi-core, clusters, and large scale grid/cloud resources. We will ensure the compatibility of the new programming model with the China Grid specifications, and will assess the viability and efficiency of the approach on a large example from the area of bioinformatics.