| Title: | What makes for worst-case performance? |
| Course: | Asyncronous Design (CPSC 538a) |
| Timeframe: | 09.2004 - 12.2004 |
| Supervisor: | Mark Greenstreet |
| Abstract: |
In asynchronous systems computational performance is determined by the speed of handshakes between
its components. Very often it is intuitively assumed that the performance of a self-timed system can
be measured by the average-case performance of its components. In ''How to Achieve Worst-Case Performance'',
Mark Greenstreet and Brian d'Alwis showed that such intuition is often wrong. They looked at regular
graphs of processors with Bernoulli processing times. They gave qualitative arguments that other networks
should show similar behaviours, but did not quantify these claims. There have been suggestions that their
results only apply to regular graphs. The goal of this project is to investigate the performance for
various network topologies, degrees of connectivity, and processing time distributions to look for
quantitative rules-of-thumb for the actual performance. The following chart summarizes the main results: |
| |
| Downloads: | Paper (PDF) [170 KB] |
minevskiy
gmail.com
|
