Measuring unique users in a billion user network is hard - accurate counting is space consuming and not easily distributable.

In this talk I will describe HyperLogLog, a probabilistic cardinality estimation algorithm and data structure and how we used it to provide breakdowns of our billion user reach.