CANCELLED - ICSA Seminar - 12/03/2020 Title: A Hybrid GraphBLAS in C++11: specification, design, implementation, and performance Abstract: We present a C++11 API to GraphBLAS, a paradigm for expressing graph computations in terms of (sparse) linear algebra, and introduce three back-ends that implement this API: a sequential reference implementation, a shared-memory parallel one using OpenMP, and a distributed-memory parallel one using the Lightweight Parallel Foundations (LPF) communications layer. The latter may be combined with the OpenMP back-end, thus resulting in a fully hybrid shared- and distributed-memory GraphBLAS implementation. We will demonstrate subset of our C++ interface, discuss key implementation choices, and point out extensions to the GraphBLAS.org C11 API we found necessary to efficiently support a distributed-memory GraphBLAS. We demonstrate the attained performance of classical graph algorithms and discuss how GraphBLAS algorithms may be encapsulated in real-life applications, even if those are implemented on top of auxiliary (Big Data) frameworks such as Spark. Bio: Albert-Jan Nicholas Yzelman was born in Zevenaar, the Netherlands and raised in Duiven. After obtaining his Atheneum diploma there in 2002, he started his scientific career at Utrecht University from which he graduated with a B.Sc. in both Mathematics and Computing Sciences, a M.Sc. in Scientific Computing, and, in 2012, a PhD for his thesis `Fast sparse matrix-vector multiplication by partitioning and reordering', written under supervision of Prof. dr. Rob H. Bisseling. Afterwards, Albert-Jan moved to KU Leuven, Belgium under Prof. Dirk Roose and Prof. Karl Meerbergen. There he worked with the Flanders ExaScience Laboratory and the ExaScience Life HPC Lab, focusing on shared-memory high-performance parallel computing mainly for sparse matrix computations. In 2015, Albert-Jan made a switch to industry to pursue a career at Huawei Technologies, where he currently holds a position as a Research Scientist and Expert. He first joined a newly set-up research group head by Bill McColl in Paris, where he continued his research on parallel computing and irregular computations with applications on scales ranging from mobile to cloud. He was promoted both in 2017 and 2019. In February 2020 he left for the Huawei Zürich Research Lab which focuses on long-term, fundamental, and open research in future computing. Mar 12 2020 14.00 - 15.00 CANCELLED - ICSA Seminar - 12/03/2020 Albert-Jan Nicholas Yzelman G.07, IF
CANCELLED - ICSA Seminar - 12/03/2020 Title: A Hybrid GraphBLAS in C++11: specification, design, implementation, and performance Abstract: We present a C++11 API to GraphBLAS, a paradigm for expressing graph computations in terms of (sparse) linear algebra, and introduce three back-ends that implement this API: a sequential reference implementation, a shared-memory parallel one using OpenMP, and a distributed-memory parallel one using the Lightweight Parallel Foundations (LPF) communications layer. The latter may be combined with the OpenMP back-end, thus resulting in a fully hybrid shared- and distributed-memory GraphBLAS implementation. We will demonstrate subset of our C++ interface, discuss key implementation choices, and point out extensions to the GraphBLAS.org C11 API we found necessary to efficiently support a distributed-memory GraphBLAS. We demonstrate the attained performance of classical graph algorithms and discuss how GraphBLAS algorithms may be encapsulated in real-life applications, even if those are implemented on top of auxiliary (Big Data) frameworks such as Spark. Bio: Albert-Jan Nicholas Yzelman was born in Zevenaar, the Netherlands and raised in Duiven. After obtaining his Atheneum diploma there in 2002, he started his scientific career at Utrecht University from which he graduated with a B.Sc. in both Mathematics and Computing Sciences, a M.Sc. in Scientific Computing, and, in 2012, a PhD for his thesis `Fast sparse matrix-vector multiplication by partitioning and reordering', written under supervision of Prof. dr. Rob H. Bisseling. Afterwards, Albert-Jan moved to KU Leuven, Belgium under Prof. Dirk Roose and Prof. Karl Meerbergen. There he worked with the Flanders ExaScience Laboratory and the ExaScience Life HPC Lab, focusing on shared-memory high-performance parallel computing mainly for sparse matrix computations. In 2015, Albert-Jan made a switch to industry to pursue a career at Huawei Technologies, where he currently holds a position as a Research Scientist and Expert. He first joined a newly set-up research group head by Bill McColl in Paris, where he continued his research on parallel computing and irregular computations with applications on scales ranging from mobile to cloud. He was promoted both in 2017 and 2019. In February 2020 he left for the Huawei Zürich Research Lab which focuses on long-term, fundamental, and open research in future computing. Mar 12 2020 14.00 - 15.00 CANCELLED - ICSA Seminar - 12/03/2020 Albert-Jan Nicholas Yzelman G.07, IF
