Section: New Results

HPC and parallelism

Participants : Dominique Lavenier, Rumen Andonov, Guillaume Chapuis, François Moreews, Charles Deltel.

• Improving time performances of Mapping quantitative trait loci (QTL) : we have developed a fast implementation of QTLMap, which takes advantage of the data parallel nature of the problem by offsetting heavy computations to a graphics processing unit (GPU). This new implementation performs up to 75 times faster than the previous multicore implementation, while maintaining the same results and level of precision . This speedup allows one to perform more complex analyses, such as linkage disequilibrium linkage analyses (LDLA) and multiQTL analyses, in a reasonable time frame. [13]

• Integration of parallelism in bioinformatics workflows: We propose a Model-Driven Architecture approach for capturing the complete design process of bioinformatics workflows. This approach is applied to graphical workflow editors and allows to quickly convert a workflow prototype in a parallel implementation. This work can have an impact on the way bioinformaticians implement their analysis and increase their productivity.[30]

• Parallel assembly on FPGAs: This research work proposes a method to reduce the overall time for assembly by using pre-processing of the short read data on FPGAs and processing its output using Velvet. We demonstrate significant speed-ups with slight or no compromise on the quality of the assembled output.[32]

• All-Pairs Shortest Paths with multi-GPU We propose a new algorithm for the All-Pairs Shortest Paths problem for graphs with good partitioning properties and its multi-GPU implementation. Our implementation targets large graphs (up to ${10}^6$ vertices) and allows graphs with negative edges to be computed. [35]