1.
Dragomir, Voichita; Stefan, Gheorghe M.
ALL-PAIR SHORTEST PATH ON A HYBRID MAP-REDUCE BASED ARCHITECTURE Journal Article
In: PROCEEDINGS OF THE ROMANIAN ACADEMY SERIES A-MATHEMATICS PHYSICS TECHNICAL SCIENCES INFORMATION SCIENCE, vol. 20, no. 4, pp. 409-415, 2019, ISSN: 1454-9069.
Abstract | BibTeX | Tags: APSP; parallelism; hybrid computation; map-reduce architecture; Floyd-Warshall algorithm
@article{WOS:000503865200012,
title = {ALL-PAIR SHORTEST PATH ON A HYBRID MAP-REDUCE BASED ARCHITECTURE},
author = {Voichita Dragomir and Gheorghe M. Stefan},
issn = {1454-9069},
year = {2019},
date = {2019-10-01},
journal = {PROCEEDINGS OF THE ROMANIAN ACADEMY SERIES A-MATHEMATICS PHYSICS
TECHNICAL SCIENCES INFORMATION SCIENCE},
volume = {20},
number = {4},
pages = {409-415},
publisher = {EDITURA ACAD ROMANE},
address = {CALEA 13 SEPTEMBRIE NR 13, SECTOR 5, BUCURESTI 050711, ROMANIA},
abstract = {All-Path Shortest-Path Problem (APSP) algorithm implemented on two
currently used architectures, a mono-core one and on a many-core one, is
compared with the implementation on Map-Reduce Architecture (MRA), a
novel many-core architecture we propose. The hybrid system using an
accelerator based on MRA is described. The system is evaluated in two
versions for miming the Floyd-Warshall APSP algorithm. A first version
is for an accelerator with the number of cores, p, exceeding the number
of vertexes, vertical bar V vertical bar, while the second uses a fix number of cells, N, vertical bar V vertical bar for = NxM, where M is a
power of 2. The programs, written for our accelerator running in
simulation, are used to evaluate the execution time for both versions.
The performance of a mono-core and of a GPU is compared with our MRA
acceleration. A 128-cell MRA engine accelerates 118x at 20 x less energy
a mono-core, and 3x at 26 x less energy a 128-core GPU.},
keywords = {APSP; parallelism; hybrid computation; map-reduce architecture; Floyd-Warshall algorithm},
pubstate = {published},
tppubtype = {article}
}
All-Path Shortest-Path Problem (APSP) algorithm implemented on two
currently used architectures, a mono-core one and on a many-core one, is
compared with the implementation on Map-Reduce Architecture (MRA), a
novel many-core architecture we propose. The hybrid system using an
accelerator based on MRA is described. The system is evaluated in two
versions for miming the Floyd-Warshall APSP algorithm. A first version
is for an accelerator with the number of cores, p, exceeding the number
of vertexes, vertical bar V vertical bar, while the second uses a fix number of cells, N, vertical bar V vertical bar for = NxM, where M is a
power of 2. The programs, written for our accelerator running in
simulation, are used to evaluate the execution time for both versions.
The performance of a mono-core and of a GPU is compared with our MRA
acceleration. A 128-cell MRA engine accelerates 118x at 20 x less energy
a mono-core, and 3x at 26 x less energy a 128-core GPU.
currently used architectures, a mono-core one and on a many-core one, is
compared with the implementation on Map-Reduce Architecture (MRA), a
novel many-core architecture we propose. The hybrid system using an
accelerator based on MRA is described. The system is evaluated in two
versions for miming the Floyd-Warshall APSP algorithm. A first version
is for an accelerator with the number of cores, p, exceeding the number
of vertexes, vertical bar V vertical bar, while the second uses a fix number of cells, N, vertical bar V vertical bar for = NxM, where M is a
power of 2. The programs, written for our accelerator running in
simulation, are used to evaluate the execution time for both versions.
The performance of a mono-core and of a GPU is compared with our MRA
acceleration. A 128-cell MRA engine accelerates 118x at 20 x less energy
a mono-core, and 3x at 26 x less energy a 128-core GPU.