ALBA CRISTINA MELO
Hakizumwami Birali Runesha is the Assistant Vice President for Research Computing and founding Director of the Research Computing Center (RCC) at The University of Chicago. He is the current President of the Great Lakes Consortium for Petascale Computation (GLCPC)-USA, member of Rwanda’s National Council of Science and Technology and founding member of Intel and Lenovo’s Project Everyscale, an exascale visionary council. Dr Runesha is a former Director of Scientific Computing and Alba Cristina Magalhaes Alves de Melo obtained her PhD in Computer Science from the Institut National Polytechnique de Grenoble (INPG), France, in 1996. Since 1997, she works at the Department of Computer Science at the University of Brasilia, Brazil, where she is now Full Professor and Vice-Head of the Department. She is also a CNPq Research Fellow level 1C. Prof. Melo is Senior Member of the IEEE Computer Society and the Chair of the IEEE Professional Chapter - Computer Society - in IEEE Center-North Region, Brazil. From 2015 to 2019, she was member of the Editorial Board of IEEE Transactions of Parallel and Distributed Systems and she is currently member of the Editorial Board of IEEE Transactions on Computers. She was Vice Program Chair of IEEE International Parallel and Distributed Processing Symposium (IPDPS) 2019 and she is co-General Chair of the IEEE International Workshop on High Performance Computational Biology (HiCOMB). She has served as Program Committee Member in many prestigious conferences such as IPDPS, Supercomputing, ICPP, ICS, Euro-Par, CCGrid, ISC, SBAC-PAD and HiPC.
Topic: Parallel Biological Sequence Comparison in GPU Clusters
Pairwise Biological Sequence Comparison is an important Bioinformatics application, which is executed thousands of times daily all over the world. The algorithms used to obtain the optimal result run in quadratic time complexity and, thus, they take a lot of time if the sequences compared are long or if there are many sequences to be compared. In order to accelerate Biological Sequence Comparison applications, GPUs have been used for more than a decade. In this talk, we first discuss works in the area of parallel biological sequence comparison with the Smith-Waterman algorithm and its variants. Then, we present CUDAlign, a fine-grained multi-GPU strategy to compared DNA sequences with up to 249 millions of characters in hundreds of GPUs. In order to achieve this, CUDAlign uses: (a) a fine-grained parallelogram-shaped strategy to exploit parallelism; (b) overlapping of computation and communication and (c) an innovative speculation technique, which is able to parallelize a phase of the Smith-Waterman algorithm that is inherently sequential. We will show that CUDAlign is able to attain the impressive rate of 10.3 TCUPS (Trillions of matrix Cells Updated per Second), the best performance for SmithWaterman in GPUs in the literature so far. We also present a pruning technique for one GPU that is able to prune more than 50% of the Smith-Waterman matrix and still retrieve the optimal alignment. Then, we discuss energy consumption of the CUDAlign tool. After this, we present our ongoing work on block pruning with multiple GPUS and show some preliminary results in an IBM Power9 + NVidia Volta platform. Finally, we discuss challenges and open research directions.