Chapter 14. Performance
Theodore Hong, Imperial College of Science, Technology, and Medicine
We live in the era of speed. Practically as a matter of course, we expect each day to bring faster disks, faster networks, and above all, faster processors. Recently, a research group at the University of Arizona even published a tongue-in-cheek article arguing that large calculations could be done more quickly by slacking off for a few months first, then buying a faster computer:
[B]y fine tuning your slacktitude you can actually accomplish more than either the lazy bum at the beach for two years or the hard working sucker who got started immediately. Indeed with a little bit of algebra we convince ourselves that there exists an optimal slack time s
In a world like this, one might well wonder whether performance is worth paying attention to anymore. For peer-to-peer file-sharing systems, the answer is a definite yes, for reasons I will explain in the next section.
Let me first emphasize that by performance, I don’t mean abstract numerical benchmarks such as, “How many milliseconds will it take to render this many millions of polygons?” Rather, I want to know the answers to questions such as, “How long will it take to retrieve this file?” or “How much bandwidth will this query consume?” These answers will have a direct impact on the success and usability of a system.
Fault tolerance is ...