Publication Details
Nikos Chrisochoides, Andrey Chernikov and Christos Tsolakis.
Poster in Publisher SPL25, Brigham and Women's Hospital & Harvard Medical School, April, 2016
Abstract
There have been numerous advances in the computer hardware over the last 15 years that drove the current and emerging architectures for exascale-era computing. Networking techniques have been integrated into the memory hierarchy resulting in multi-layered (w.r.t. memory and network) supercomputing architectures. In the past 15 years we developed the theory for parallel mesh generation capable of taking advantage of multi-layered architectures. y combining our experience in designing algorithms and software, for each of the layers, we have developed a novel telescopic approach which explores concurrency at different levels of granularity: (1) medium-grain level for multiple cores within a single chip, and (2) coarse-grain level for multiple chips within a node or a cluster, both using proper error metrics and application-specific discrete data and continuous domain decomposition methods.
