Most models work well if you can include some information about workloads or transactions, because then you can relate the results to the service levels seen by individual users. However, this is not always necessary. Suppose you have a server running a single main workload - a good example would be a Windows NT server supporting SQL Server or IIS. You may be asked to determine how much more of the same workload the server can support. Even if you don't know exactly how many transactions per hour are being processed, you can still answer the question by building a "system-only" model, which treats the whole workload as a single unit. In fact, you can build "system-only" models to represent servers supporting several workloads - all you need to do is express the approximate split of CPU use across the workloads. This information is usually available, or can be derived from, standard system monitors such as Windows NT Performance Monitor or the UNIX sar and ps commands. A specialist modeling tool such as Metron's Athene Planner lets you build system-only models quickly and easily.