Here is a useful link with several presentations from Intel's
D. Levinthal about the Core i7 micro-architecture and how to
make use of the PMU to solve performance bottlenecks.
Of particular interest to people asking about which events are useful
or about the methodology to track down problems, I recommend you
look at the first presentation on the page (SW optimization on Core i7).
Thanks to CERN for providing the presentations online.