Continuous Remote Analysis for Improving Distributed Systems ' Performance
|
Abstract:
Engineering a highly distributed system requires the ability to evaluate and optimize the protocols that control the movement and processing of information throughout the system. Because the design of such protocols is often characterized by conflicting goals and trade-o#s, the designer must calibrate the parameters of the protocols, seeking the best balance of performance in the most common usage scenarios. Unfortunately, fully testing these calibrations requires experiments conducted on large, expensive testbeds that are very di#cult to deploy and maintain. In this paper, we propose a new approach for the optimization of a highly distributed system's performance. The approach is based on leveraging data collected from fielded components to fine-tune the behavior of the system and its protocols. Captured data is "replayed " in simulations performed directly in the field during o#-peak hours. The results of these simulations are then used to control the system directly in the field, and/or to report aggregate performance and behavior information to the system designer. 1.
Citations
| 432 | Design and evaluation of a wide-area event notification service – Carzaniga, Rosenblum, et al. |
| 43 | Gamma System: Continuous Evolution of Software after Deployment – Orso, Liang, et al. - 2002 |
| 38 | Monitoring Deployed Software Using Software Tomography – Bowring, Orso, et al. - 2002 |
| 24 | Visualization of program-execution data for deployed software – Orso, Jones, et al. - 2003 |
| 14 | A benchmark suite for distributed publish/subscribe systems – Carzaniga, Wolf - 2002 |
