Using buffer cache inspection for sizing feedback
The recent examples give you the basic two things to look for when deciding if your shared_buffers cache is large enough. First, compare what percentage of the relations you believe are important to your application's performance appear to be cached. If this number is low, you may benefit from a larger buffer cache. A helpful secondary look at this information is available if you combine this with hit rates from views such as pg_stat_user_tables, which Chapter 11, Database Activity and Statistics.
The balance of popular (high usage count) versus transient (low usage count) pages tells you a lot about whether your cache is sized appropriately. If most of your pages have low usage counts (0,1), but you're still getting good hit rates from the table statistics, you can probably decrease the size of the buffer cache without a performance loss. Even the simplest operating system LRU algorithm is capable of making good caching decisions...
