Requests arrive at random and wait in line whenever every server is busy. Because arrivals bunch up, a line forms long before the servers are full, and the wait grows like 1/(1 − ρ): a server 90% busy keeps requests waiting nine times as long as it serves them. The slowest requests, the tail, suffer most.