A method of scheduling one or more virtual machines in a cloud computing infrastructure may include identifying, by a computing device, lag time data that has been collected over a period of time and that corresponds to one or more virtual machines in a cloud computing infrastructure, computing, by the computing device, a cumulative description of the identified lag time data, identifying a target performance level, determining, by the computing device, an estimated lag time that corresponds to the target performance level, and determining, by the computing device, whether to start a new virtual machine based, at least in part, on the estimated lag time.