Using fuzzy control to maximize profits in service level management

  • Authors:
  • Y. Diao;J. L. Hellerstein;S. Parekh

  • Affiliations:
  • IBM Research Division, Thomas J. Watson Research Center, Yorktown Heights, New York;IBM Research Division, Thomas J. Watson Research Center, Yorktown Heights, New York;IBM Research Division, Thomas J. Watson Research Center, Yorktown Heights, New York

  • Venue:
  • IBM Systems Journal
  • Year:
  • 2002

Quantified Score

Hi-index 0.00

Visualization

Abstract

The growth of e-commerce is creating demand for services with financial incentives for service providers. Such services are specified by service level agreements (SLAs) in which revenue accrues with the number of completed transactions, and costs are incurred if response times go above a specified threshold. In this paper we propose a profit-oriented feedback control system that automates the admission control decisions in a way that balances the loss of revenue due to rejected work against the penalties incurred if admitted work has excessive response times. One approach to making these trade-offs is to employ classical control theory (e.g., proportional integral controllers); however, doing so requires a labor-intensive design process that tailors the controller to a specific workload and profit model. Instead, we develop a fuzzy control algorithm that implements hill climbing logic to maximize profits and handle the stochastics that make profits quite "bumpy." Our studies of a Lotus Notes® e-mail server reveal that the fuzzy controller requires little human intervention and is robust to changes in workloads and values of SLA parameters. Further, we prove that our design of the fuzzy controller leads to a value of the admission control parameter that maximizes steady state profits.