Title: MADRE: The Memory-Aware Data Redistribution Engine
Speaker: Stephen Siegel
Department of Computer & Information Sciences, University of Delaware
Joint work with Andrew Siegel, Argonne National Laboratory.
Abstract:
A new class of algorithms is presented for efficiently carrying out
many-to-many parallel data redistribution in a memory-limited
environment. Key properties of these algorithms are explored, and
their performance is compared using idealized benchmark
problems. These algorithms form part of a newly developed MPI-based
library MADRE (Memory-Aware Data Redistribution Engine), an open
source toolkit designed for easy integration with application codes to
improve their performance, portability, and scalability.