CMU-CS-99-138
Computer Science Department
School of Computer Science, Carnegie Mellon University



CMU-CS-99-138

An Extensible Toolkit For Resource Prediction In Distributed Systems

Peter A. Dinda, David R. O'Hallaron

July 1999

CMU-CS-99-138.ps
CMU-CS-99-138.pdf


Keywords: Resource prediction, performance prediction, linear time series models, time series prediction, prediction toolkits, distributed systems, predictive scheduling, application-level scheduling


This paper describes the design, implementation, and performance of RPS, an extensible toolkit for building flexible on-line and off-line resource prediction systems in which resources are represented by independent, periodically sampled, scalar-valued measurement streams. RPS-based prediction systems predict future values of such streams from past values. Systems are composed at run-time out of an extensible set of communicating prediction components which are in turn constructed using RPS's sensor, prediction, and communication libraries. We have used RPS to evaluate predictive models and build on-line prediction systems for host load and network bandwidth. The overheads involved in such systems are quite low.

36 pages


Return to: SCS Technical Report Collection
School of Computer Science homepage

This page maintained by reports@cs.cmu.edu