For an effective operating system, deadlock should be checked every time a resource request is made. This approach ensures immediate detection when processes request resources, preventing the deadlock from persisting or causing system slowdown. Although this method incurs the most computational overhead, it is the most comprehensive in timely detecting and managing deadlocks. Alternatively, some systems may check at fixed intervals or when certain system performance clues appear, but these methods risk delayed detection and more complex recovery.