SESAME: Scalable Systems Software Measurement and Evaluation
Abstract
The goal of the SESAME project has been to create a set of tools to study the performance of operating systems. This project developed a method for observing component behavior inside the operating system kernel in fully running computers, a method for modeling the effects of changes, and a method for sharing the results among researchers. These have been fully implemented and tested on a massively parallel processing system (MPP). The project has used this implementation to identify details of system operation that were both contrary to documentation, and difficult to establish from user-level observation. In the process, two especially important lessons have been learned. The first is that extracting data from an OS kernel is a very challenging task. Although placing "diagnostic code" into an OS seems simple enough, to be useful the method must handle the large volumes of data produced without degrading system performance, organize it in a consistent analyzable way, and make the process flexible and extensible enough to allow arbitrary investigations. The method and programming described are substantially portable, and other teams will almost certainly find it much more practical to adapt them, than to undertake a three-year process of reinvention from scratch. The second important lesson is that such invasive, detailed tools are likely best used by inside members of OS development teams, or their quality assurance counterparts. In short, exposing such intimate system behavior can expose trade secrets. The measurement team was bound by a non-disclosure agreement with the vendor, and faced a constant dilemma between its obligations not to disclose protected intellectual property, and pressure from analysts for detailed information.
Document Details
- Document Type
- Technical Report
- Publication Date
- Aug 01, 2000
- Accession Number
- ADA385148
Entities
People
- Gerald Popek
- Leonard Kleinrock
- Rajive Bagrodia
Organizations
- University of California, Los Angeles