An automated method of managing computing resources having a workload of a given type comprises providing resource data collectors for collecting data regarding performance of the resources, such as a server network, in accordance with the type of workload; developing a forecast of utilization of the resources, based on historical performance data; and collecting real-time performance data regarding the resources running under the workload. The method then includes analyzing the performance data and the forecast to identify a critical resource: and automatically adjusting a capacity of the resource to provide steady-state performance of the resource under the workload. Additional hardware resources such as CPUs, computer memory and computer disk storage are initially available to, but unused by, the computing resources.