Design of fast collective communication functions on clustered workstations with ethernet and myrinet

Authors:
Dongyoung Kim;Dongseung Kim
Affiliations:
Department of Electrical Engineering, Korea University, Seoul, Republic of Korea;Department of Electrical Engineering, Korea University, Seoul, Republic of Korea
Venue:
ISHPC'05/ALPS'06 Proceedings of the 6th international symposium on high-performance computing and 1st international conference on Advanced low power systems
Year:
2005

Citing 7
Cited 0

Two algorithms for barrier synchronization

International Journal of Parallel Programming
PVM: Parallel virtual machine: a users' guide and tutorial for networked parallel computing

PVM: Parallel virtual machine: a users' guide and tutorial for networked parallel computing
Optimization of MPI collectives on clusters of large-scale SMP's

SC '99 Proceedings of the 1999 ACM/IEEE conference on Supercomputing
Advanced Computer Architecture: Parallelism,Scalability,Programmability

Advanced Computer Architecture: Parallelism,Scalability,Programmability
MPI and Embedded TCP/IP Gigabit Ethernet Cluster Computing

LCN '02 Proceedings of the 27th Annual IEEE Conference on Local Computer Networks
On optimizing collective communication

CLUSTER '04 Proceedings of the 2004 IEEE International Conference on Cluster Computing
Fast broadcast by the divide-and-conquer algorithm

CLUSTER '04 Proceedings of the 2004 IEEE International Conference on Cluster Computing

Quantified Score

Hi-index	0.00

Visualization

Abstract

In cluster computing, current communication functions under MPI library are not well optimized. Especially, the performance is worse if there are multiple sources and/or destinations involved, which are the cases of collective communication. Our algorithms uses multidimensional factorization and pairwise exchange communication/dissemination methods to improve the performance. They deliver better performance than previous algorithms such as ring, recursive doubling and dissemination algorithms. Experimental results show the improvement of 50% or so over MPICH version 1.2.6 on a Linux cluster.