Here's a fascinating (for certain, functionally pathetic definitions of fascination common to geeks like me) look inside Google's query serving architecture. It takes a look at how they take advantage of parallelism (at the network, hardware, and processor levels) inherent in serving search engine queries and explain their practical choices based on hardware costs, power consumption, and the savings possible when using cheap hardware with fault-tolerant software. (via Aaron Swartz's Google weblog)
What you don’t know won’t help you much either.
— D. Bennett