[ChuckMcM]

This is a really important topic if you're doing web services. Especially if you're doing parallel processing services. When I was at Blekko the "interesting" queries were the ones above the 95th percentile because they always indicated "something" that hadn't worked according to plan. Sometimes it was a disk going bad on one of the bucket servers, sometimes it was a network port dropping packets, and sometimes it was a corrupted index file. But it was always something that needed to be looked at and then (usually) fixed.

It also always separated the 'good' Ad networks from the 'bad' ones as the bad ones would take to long to respond.