EN FR
EN FR


Section: New Results

HPC Component Model

Participants : Hélène Coullon, Vincent Lanore, Christian Perez, Jérôme Richard.

3D FFT and L2C

We have studied the relevance of dealing with 3D FFT parallel algorithms with the software component model L2C  [31] . We have implemented several existing 3D FFT algorithms, and we have evaluated their performance, their scalability, and their reuse rate. Experiments made on clusters of Grid'5000 and on the Curie supercomputer up to 8192 cores show that L2C based 3D assemblies are scalable and have the same kind of performance than existing 3D libraries such as FFTW or 2DECOMP. This work confirms than components can be used for optimized HPC applications

Stencil Skeletons in L2C

Mesh-based scientific simulation is an important class of scientific application which could benefit from component models. Therefore, we have studied and designed a first adaptation of the SIPSim model  [33] (Structured Implicit Parallelism for scientific Simulations) to handle HPC component models. The heat equation application has been implemented on top of L2C following this adapted SIPSim model. First experiments on clusters of Grid'5000 and on the Curie supercomputer show promising results, of which a complete analysis is still ongoing. This work is a first step toward a complete implicit parallelism stencil skeleton using L2C.

Reconfigurable HPC component model

High-performance applications whose structure changes dynamically during execution are extremely complex and costly to develop, maintain and adapt to new hardware. Such applications would greatly benefit from easy reuse and separation of concerns which are typical advantages of component models. Unfortunately, no existing component model is both HPC-ready (in terms of scalability and overhead) and able to easily handle dynamic reconfiguration.

We aim at addressing performance, scalability and programmability by separating locking and synchronization concerns from reconfiguration code. To this end, we have defined directMOD, a component model which provides on one hand a flexible mechanism to lock subassemblies with a very small overhead and high scalability, and on the other hand a set of well-defined mechanisms to easily plug various independently-written reconfiguration components to lockable subassemblies. We evaluate both the model itself and a C++/MPI implementation called directL2C based on L2C.