Hierarchial Modeling of Supercomputer Reliability: Abstract
Statistical Engineering Division
Seminar Series
Hierarchial Modeling of SUpercomputer Reliability: Abstract
Kenneth Ryan
Department of Statistics
Iowa State University
Los Alamos National Laboratory is home to one of the World's
fastest supercomputers--Blue Mountain. This machine was
created by parallelizing desktop computers. To help determine
whether or not this type of architecture represents the
future of supercomputing, reliability must be estimated.
This paper presents and analyzes failure data for Blue Mountain.
Nonhomogeneous Poisson processes are fit to the data within a
hierarchical Bayesian framework. The task of selecting
hyperparameters is discussed, and Bayes factors are used to
compare models.
Date created: 6/5/2001
Last updated: 6/21/2001
Please email comments on this WWW page to
sedwww@nist.gov.