I'm not one to peddle my own wares, but Alchemy (distributed memory array storage) fits the bill for a slightly more elegant solution. Think of a hash with key

I'm not one to peddle my own wares, but Alchemy (distributed memory array storage) fits the bill for a slightly more elegant solution. Think of a hash with key => array pairs. With Alchemy, you can just stuff the times of your most recent "actions" into a key. Here's a Ruby psuedocode example: alchemy[ratelimit_72.26.203.98_2009-01-07] << 1.minute.ago alchemy[ratelimit_72.26.203.98_2009-01-07] << 20.seconds.ago alchemy[ratelimit_72.26.203.98_2009-01-07] << 5.seconds.ago alchemy[ratelimit_72.26.203.98_2009-01-07].select{ |e| e > 30.seconds.ago }.size In that brief example we've defined a true rolling window (30 seconds) with the ability to grab the true count of actions performed (size) with only one cache query. And similar to the pure memcached version the article suggests, you don't have issues with race conditions.