This video is from our good friends at USENIX's recent LISA conference in Boston.

"Linux Systems Capacity Planning: Beyond RRD and top", by Rodrigo Campos

As infrastructure costs rise, there's an urgent need to squeeze more performance from the same hardware. After several years of measuring and managing the capacity of thousands of Linux servers, we have learned that most typical tools and metrics are not sufficient to predict performance bottlenecks, particularly during traffic spikes. By using queue theory formulas and instrumenting our applications we were able to find the limits of our systems, improve reliability, and maximize throughput and performance.