About the paper
| Title |
Dataflow of Matrix Multiplication Algorithm through Distributed Hadoop Environment |
| Book |
ICIST 2016 Proceedings |
| Pages |
46-49 |
| Abstract: Increasing of processors' frequencies and
computational speed with components scaling is slowly
reaching its saturation with current MOSFET technology.
From today's perspective, the solution lies either in further
scaling in nanotechnology, or in parallel and distributed
processing. Parallel and distributed processing have always
been used to speedup the execution further than the current
technology had been enabling. However, in parallel and
distributed processing, dependencies play a crucial role and
should be analyzed carefully. The goal of this paper is the
analysis of dataflow and parallelization capabilities of
Hadoop, as one of the widely used distributed environment
nowadays. The analysis is performed on the example of
matrix multiplication algorithm. The dataflow is analyzed
through evaluation of the execution timeline of Map and
Reduce functions, while the parallelization capabilities are
considered through the utilization of Hadoop's Map and
Reduce tasks. The implementation results on 18-nodes
cluster for various parameter sets are given. |
Full citation
Ćirić, V.,
Živanović, F.,
Stojanović, N.,
Milovanović, E.,
Milentijević, I.
Dataflow of Matrix Multiplication Algorithm through Distributed Hadoop Environment. In:
Konjović, Z.,
Zdravković, M.,
Trajanović, M.
(Eds.)
ICIST 2016 Proceedings, pp.46-49, 2016
Keywords
Parallel and distributed processing