Evaluation of interconnection network performance under heavy non-uniform loads

Authors:
C. Izu;J. Miguel-Alonso;J. A. Gregorio
Affiliations:
Department of Computer Science, The University of Adelaide, SA, Australia;Dep. of Computer Architecture and Technology, The University of the Basque Country, San Sebastian, Spain;Computer Architecture Research Group, Universidad de Cantabria, Santander, Spain
Venue:
ICA3PP'05 Proceedings of the 6th international conference on Algorithms and Architectures for Parallel Processing
Year:
2005

Citing 8
Cited 3

NIFDY: a low overhead, high throughput network interface

ISCA '95 Proceedings of the 22nd annual international symposium on Computer architecture
A Necessary and Sufficient Condition for Deadlock-Free Routing in Cut-Through and Store-and-Forward Networks

IEEE Transactions on Parallel and Distributed Systems
The adaptive bubble router

Journal of Parallel and Distributed Computing
The Alpha 21364 Network Architecture

IEEE Micro
The Measured Network Traffic of Compiler-Parallelized Programs

ICPP '02 Proceedings of the 2001 International Conference on Parallel Processing
On the Design of a High-Performance Adaptive Router for CC-NUMA Multiprocessors

IEEE Transactions on Parallel and Distributed Systems
Principles and Practices of Interconnection Networks

Principles and Practices of Interconnection Networks
The Case of the Missing Supercomputer Performance: Achieving Optimal Performance on the 8,192 Processors of ASCI Q

Proceedings of the 2003 ACM/IEEE conference on Supercomputing

Throughput fairness in k-ary n-cube networks

ACSC '06 Proceedings of the 29th Australasian Computer Science Conference - Volume 48
A new congestion control algorithm for improving the performance of a broadcast-based multiprocessor architecture

Journal of Parallel and Distributed Computing
On-Chip Network Evaluation Framework

Proceedings of the 2010 ACM/IEEE International Conference for High Performance Computing, Networking, Storage and Analysis

Quantified Score

Hi-index	0.00

Visualization

Abstract

Many simulation-based performance studies of interconnection networks are carried out using synthetic workloads under the assumption of independent traffic sources. We show that this assumption, although useful for some traffic patterns, can lead to deceptive performance results for loads beyond saturation. Network throughput varies so much amongst the network nodes that average throughput does not reflect anymore network performance as a whole. We propose the utilization of burst synchronized traffic sources that better reflect the coupled behavior of parallel applications at high loads. A performance study of a restrictive injection mechanism is used to illustrate the different results obtained using independent and non-independent traffic sources.