Performance Modeling of the Blue Gene Architecture

Authors:
Darren J. Kerbyson;Adolfy Hoisie
Affiliations:
Los Alamos National Laboratory, NM;Los Alamos National Laboratory, NM
Venue:
JVA '06 Proceedings of the IEEE John Vincent Atanasoff 2006 International Symposium on Modern Computing
Year:
2006

Citing 0
Cited 3

ScalaExtrap: trace-based communication extrapolation for spmd programs

Proceedings of the 16th ACM symposium on Principles and practice of parallel programming
ScalaExtrap: Trace-based communication extrapolation for SPMD programs

ACM Transactions on Programming Languages and Systems (TOPLAS)
Dataflow-driven GPU performance projection for multi-kernel transformations

SC '12 Proceedings of the International Conference on High Performance Computing, Networking, Storage and Analysis

Quantified Score

Hi-index	0.00

Visualization

Abstract

In this work we analyze the performance of the largest installation of the Blue Gene/L architecture. This currently has the highest peak performance of any supercomputer. The performance characteristics of this machine are considered that gives rise to achievable application performance. By using a detailed performance model of Sweep3D - a kernel application of interest to Los Alamos National Laboratory we verify that the performance of Blue Gene/L is as expected. In addition the model is used to analyze a possible configuration of a peta-flop version of Blue Gene which contains an increased number of processor cores operating at a higher clock frequency. Uncertainties in the performance characteristics of this future system are captured through a sensitivity analysis on the achievable processing rate, communication latency, and bandwidth. The analysis shows the expected level of performance for Sweep3D that may be achieved on such a system in comparison to current systems.