Section: New Results

Parallelism

Processing-in-Memory

Participants : Charles Deltel, Dominique Lavenier.

The concept of PIM (Processor In Memory) aims to dispatch the computer power near the data. Together with the UPMEM company (http://www.upmem.com/), which is currently developing a DRAM memory enhanced with computing units, we investigate the parallelization of two bioinformatics algorithms for this new type of memory: sequence alignment and mapping [34] [33]. The first results show that blast-like algorithms or mapping algorithms can highly benefit from such memory and speed-up of more than 25 can be achieved [26].

GPU for graph algorithms

Participants : Rumen Andonov, Dominique Lavenier.

We describe three algorithms and their associated GPU implementations for two types of shortest path problems. These implementations target computations on graphs with up to millions of vertices and executions on GPU clusters. The first two algorithms solve the All-Pairs Shortest Path (APSP) problem. The first of these two algorithms allows computations on graphs with negative edges while the second trades this ability for better parallel scaling properties and improved memory access. The third algorithm solves the Single-Pair Shortest Path (SPSP) query problem. Our implementations efficiently exploit the computational power of 256 GPUs simultaneously. All shortest paths of a million vertex graph can be computed in 6 minutes and shortest path queries on the same graph are answered in a quarter of a millisecond. These implementations proved to be orders of magnitude faster than existing parallel approaches[30].