A method, apparatus, system, and signal-bearing medium that, in an embodiment, calculate the average number of servers that executed an application in a previous time period, decide whether the current number of servers that execute the application is less than the average number of servers, and, if so, power on the average number of servers. The application is then distributed across the average number of servers. In an embodiment, the server is powered on by increasing the clock speed of the processors associated with the servers. In an embodiment, a determination is periodically made whether the response time of the application is greater than a threshold, and, if so, the servers are powered on and the application is distributed across the servers. In an embodiment, a determination is periodically made whether the response time of the application is less than a threshold, and, if so, servers are powered off and the application is distributed across the remaining servers.