SimBA: A discrete-event simulator for performance prediction of volunteer computing projects

David A Flores, University of Texas at El Paso

Abstract

Volunteer Computing (VC) uses computing resources connected through the Internet and owned by the public to address fundamental problems in science. VC environments provide scientific projects with higher throughput than traditional computing systems such as clusters and supercomputers at a lower cost in terms of installation, maintenance, power, and infrastructure. However, these environments are particularly challenging because of the nature of their resources: they are volatile, error prone, and heterogeneous. ^ In order to improve the performance of VC environments, project designers need to study the efficiency of different project parameters and scheduling policies without affecting the VC community. Several challenges are associated with such efforts, when performed within VC environments. First, the time required to measure project throughput under different parameter settings can be significant. Second, every experiment is unique and unrepeatable. Third, problems associated with testing might upset volunteers. Finally, experiments are limited to the actual available resources. ^ The need for mechanisms to help efficiently design VC projects and the challenge of testing new strategies on in-production VC environments are the main forces that have driven us to design and implement SimBA, a discrete-event simulator that accurately models the main functions of VC environments. SimBA simulates the generation, distribution, collection, and monitoring of tasks executed under the Berkeley Open Infrastructure for Network Computing (BOINC), a well-known, master-worker, runtime framework used in VC projects. ^ The contributions of this thesis are as follows: (1) design and development of SimBA; (2) demonstration that SimBA can be easily extended to include new features and scheduling policies; (3) experimental results that show that SimBA's performance predictions are within approximately 5% of the performance reported by two BOINC applications; and (4) examples of how SimBA can and has been used to experiment with VC environments.^

Subject Area

Computer Science

Recommended Citation

Flores, David A, "SimBA: A discrete-event simulator for performance prediction of volunteer computing projects" (2007). ETD Collection for University of Texas, El Paso. AAI1444125.
http://digitalcommons.utep.edu/dissertations/AAI1444125

Share

COinS