I ran across this gist on Hacker
News which has a quick rundown of the
latencies of various operations. It’s nothing earth shattering, but if
you compare the intra-data center round-trip (0.5ms) to reading 1MB of
data from disk (20ms) you can pretty quickly see why systems like
memcached work so well even after adding the overhead of reading from
memory (0.25ms) and some time for the sending and receiving of the
request. Summed up the memcached response is often back in a few ms,
nearly an order of magnitude faster than reading from local disk.