|
Title:
|
A HYBRID PARALLEL VERSION OF ICTM FOR CLUSTER OF NUMA MACHINES |
|
Author(s):
|
Mateus Raeder, Neumar Ribeiro, Dalvan Griebler, Luiz Gustavo Fernandes, Márcio Castro |
|
ISBN:
|
978-989-8533-06-7 |
|
Editors:
|
Hans Weghorn, Leonardo Azevedo and Pedro Isaías |
|
Year:
|
2011 |
|
Edition:
|
Single |
|
Keywords:
|
Hybrid Programming, MPI, OpenMP, NUMA, Memory Affinity, Cluster |
|
Type:
|
Full Paper |
|
First Page:
|
291 |
|
Last Page:
|
298 |
|
Language:
|
English |
|
Cover:
|
|
|
Full Contents:
|
click to dowload 
|
|
Paper Abstract:
|
Recently, the emergence of new technologies enabled the creation of clusters with multiprocessor nodes. In these architectures, computing nodes are composed by more than one processor or core sharing the same memory. In some cases, processes placed on different processors or cores may have different response time when accessing memory. A cluster composed by such nodes is said to be a cluster of Non-Uniform Memory Access (NUMA) machines. In this scenario, parallelism can be better extracted using hybrid programming (MPI/OpenMP) along with memory affinity policies. In this work we propose a hybrid parallel version with memory affinity of the Interval Categorizer Tessellation Model (ICTM) for a cluster of NUMA machines. Our results show that the hybrid parallelization of ICTM allows better scalability and that we achieved speed-ups up to 40 while combining OpenMP, MPI and memory affinity. |
|
|
|
|
|
|
