High Performance Cluster Computing

Multivariate Analyses with Large Datasets

Parallel computing has gained considerable notoriety in the high performance computing arena. One form of lowly parallel computing, specifically that of cluster computing, has emerged as popular due in part to the technological advances in personal computer hardware over the past two decades. In addition to this heightened performance, the cost of these machines have decreased, making clusters an economically advantageous replacement for the traditional supercomputer. Finally, the introduction of widely available parallel tools (such as the Message Passing Interface or Parallel Virtual Machine) insured that the architecture required little in the way of development and programming cost for usage. Vanessa Beddo's research involves the parallel implementation (and evaluation) of various statistical algorithms, ranging from the elementary calculation of descriptive statistics to more advanced procedures. Code will soon be made available online (as this page is currently under construction), but you will also have contact me so that I may mail you the instructions as well as the program you desire. All procedures have been written in C and MPI. They have been developed on the departmental cluster www.stat.ucla.edu/cluster. You may reach Vanessa via email at VBeddo@stat.ucla.edu with any requests or inquiries.

Analysis of California Adolescent Tobacco Data

This is an analysis of the California Healthy Kids Survey, focusing on the tobacco usage of adolescents. Data has been collected from California high school students during years 1998 through 2001, resulting in 607,007 surveys. Each survey may contain up to seven components, producing up to 300 variables of interest. Due to the size of this dataset, widely used statistical programs fail to complete many of the required analyses. Parallel programming is employed for data manipulation, imputation and analysis. Implementation and development of these programs take place on the departmental cluster www.stat.ucla.edu/cluster. Ongoing research in the Division Cancer Prevention and Control Research of the Jonsson Comprehensive Cancer Center at UCLA.
Investigators: Vanessa Beddo and Coen Bernaards.

UCLA Department of Statistics
Last updated: 26-Jun-2002
Access count is: 62587, since 02-Oct-2003
Maintained by: V. Beddo and C. Bernaards [webstaff@stat.ucla.edu]