That roughly a little under 100K requests per second. Things to note: Do you perform frequent writes to the DB? These will invalidate the page cache for all open connections, potentially slowing subsequent selects. If not, consider that the default page cache size is anemic (~10MB), consider increasing that, substantially. Also did you consider mmap? This can ensure more of the db is held in system memory despite the writes and can prevent expensive trips to the actual storage media. You might want to try larger values, with and without page cache under your workload to see what works. I am hazarding a guess that mmap will yield the highest benefit in your particular case.