Seminarium: Systemy Rozproszone
13 marca 2025 12:15, sala 4070
Mikołaj Komar
Effective implementation of Sparse Matrix as a Data Structure for Distributed Ranges library
Sparse matrices have many uses, ranging from network theory to solving partial differential
equations. However, because of their size, the operations on them can be time-consuming and
inefficient.
The Distributed Ranges Library aims to simplify the use of distributed containers
by using new features from C++20, such as ranges, to provide a high-level API that provides
the power of distributed computing without having to delve into low-level details.
Today I present an efficient implementation of a distributed sparse matrix data structure
that satisfies the distributed range concept of the Distributed Ranges Library.
This allows it to be used with all algorithms that support its concept-based architecture.
Its performance
is measured by showing its strong and weak scaling properties using GPU-based NUMA
architecture and sparse matrices originating from online sources, such as the SuiteSparse
Matrix Collection. Zapraszam,
Mikołaj Komar
Bibliografia: