When no tasks are ongoing, nothing of importance is occurring, so the best speed to use is generally the minimum available. However, since our inference of tasks is imperfect, there may be ongoing tasks even when RightSpeed believes there are no such tasks. For instance, a task may have been triggered by a timeout instead of by a user interface event. We deal with this by reverting to a traditional interval-based scheduler when we know of no ongoing tasks. Such a scheduler divides time into intervals of some fixed length and chooses a speed for each interval based on the CPU utilization of recent past intervals. This way, if the CPU becomes busy from working on a task we cannot detect, the interval-based scheduler will nevertheless increase speed to deal with this unknown work.
One caveat is that when the number of known tasks becomes zero, recent past CPU utilization will likely be high because the system just finished working on a task. RightSpeed knows that this recent utilization is a poor predictor of future CPU utilization because it reflects a task that is no longer active. However, an interval-based scheduler has no knowledge of tasks, so it will interpret the high recent utilization as a sign that the next intervals will have high utilization. Accordingly, it will use an unnecessarily high CPU speed. To prevent this problem, when the number of known tasks becomes zero, RightSpeed waits for a short period of time at the minimum CPU speed before initiating the interval-based scheduler.