Computer Science Department
School of Computer Science, Carnegie Mellon University


An Adaptive Threshold-Based Policy
for Sharing Servers with Affinities

Takayuki Osogami, Mor Harchol-Balter, Alan Scheller-Wolf*, Li Zhang**

January 2004

Keywords: Load sharing, threshold policies, affinities, adaptive, multiserver systems, dimen-sionality reduction, busy period transitions, cµ rule

We evaluate the performance of threshold-based job allocation policies in a heterogeneous distributed computing system, where servers may have different speeds, and jobs may have different service demands, importance, and/or affinities for different servers. We find that while threshold-based policies typically yield low mean response time, these policies are not robust with respect to fluctuations or misprediction of the load. We propose a new adaptive dual-threshold policy and show that this policy yields low mean response time while also being robust.

21 pages

*Graduate School of Industrial Administration, Carnegie Mellon University
**IBM Research, Thomas J. Watson Research Center

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

This page maintained by