About me
Advisory Software Engineer
Research lab: Watson Research Center (Hawthorne)
I joined IBM Research in 1999. Prior to that, I attended graduate school at the University of Washington.
At IBM, my major contributions have been as a member of the (GAC) team, within the Adaptive Systems department. GAC is aimed at addressing the challenge of managing today's complex, multi-tier, distributed, heterogenous enterprise systems. For more on this challenge, see the Autonomic Computing initiative. Our group focuses on the performance management and performance tuning aspects of this problem. Many enterprises that would like to utilize today's state-of-the-art IT tools and infrastructure are faced with the frustrating and expensive problem of having to employ a cadre of highly skilled humans who are experienced in the black art of getting these systems to work to a reasonable approximation of their maximum potential. GAC is about making these systems more autonomous by incorporating self-tuning technology. This reduces the need for such expertise, and also reduces the inefficiencies that arise from possible misconfigurations of the system. Thus, we aim to lower the cost-of-ownership of these complex systems.
My work has focused on using techniques from the field of Control Theory to build reactive control systems for online tuning of the parameters of computing systems. Control systems are prevalent in a wide variety of complex physical systems (airplanes, nuclear plants, etc) so it seems reasonable the same techniques should work for computer systems. This is certainly not a new idea - control theoretic ideas are commonly used in networking and multimedia systems. However, they have been less used higher up in the application stack -- especially in the context of performance tuning of middleware and servers. To date, we have had good success in managing performance in a variety of systems, including Lotus Notes (Domino), Apache and most recently, DB2. Working in close collaboration with colleagues from the IBM Toronto Lab, we have added autonomic features for both the v8 and v9 releases of DB2. In addition to control theory, we are also using ideas from Statistics, Machine Learning and Queueing Systems.
Based on our experiences in researching and applying basic control techniques for performance management, my colleagues & I have written a book "Feedback Control of Computing Systems". This is an introductory textbook that covers the basics of discrete-time control theory. The main distinguishing factor from other introductory texts, is that all the examples in our book are about computing systems (Lotus Notes, Apache, etc) and the focus is on application, rather than a deep explanation of the theory or mathematics.
Current Projects:
- Cross-tier Performance Management for Websphere and DB2
Past Projects:
- DB2 Self-tuning Memory (released in DB2 UDB v9)
- Throttling Utilities in DB2 (released in DB2 UDB v8.1, enhanced in v8.2)
- Controlling multiple parameters in Apache to regulate CPU/Memory Utilization, Response Time, etc
- Admission control in Lotus Notes for regulating service delays.
Last updated 11 Jul 2006