SD can be used as you describe. If I were modeling CPU usage, I would use "CPU cycles in use" (the actual capacity) of the CPU as the accumulation measure. In this case, the accumulation is like a velocity (cycles per second being used). Its inflow is acceleration (cycles per second per second) and its outflow is deceleration (cycles per second per second). The CPU flow could be a co-flow with the various flows that use CPU cycles (operating system daemons running, or user programs running). Each new process entering execution (inflow to user processes running) would have an average number of CPU cycles required. These additional CPU cycles would be calculated as a part of the inflow to CPU cycles in use. If there are significantly different processes with different average CPU cycles required, then these can be captured as separate co-flows (system daemons versus user processes).