Analyzing buffer cache contents
You've seen and been told how using a block will increase its usage count. You've also seen and been told how a dirty block might make its way out to disk. This wasn't intended just as an academic exercise. Believe it or not, all of this information is useful for determining how large your shared buffer cache should be!
If you want to do better than following a rule of thumb for how big to set shared_buffers relative to the OS cache, you have two options. You can run your own benchmarks with your application and see how the results vary, based on the amount of shared memory dedicated to the database. Just be careful to account for the influence of the larger OS cache when running multiple such tests, or you can use knowledge of how the buffer cache works from inside it to help make that decision.
