In a system, alarms might be managed at different levels and by different applications. An alarm might be detected in an instrument, but the full alarm model for that alarm instance might be maintained in a higher level Server. This can cause issues in cases where the instrument is restarted or the UA Server that is acting as an alarm server is restarted. It is desirable that the state of alarms is maintained and that after any restart of either application all alarms recover their same state as before the restart. But it is acceptable if an alarm might require some management actions again following a restart. It is not acceptable that a device that is in alarm is no longer reported as being in alarm. For example, if the AlarmManageris restarted when it recovers it is able to detect that the device has an alarm, but it might not be able to recover that the alarm has a comment that had been entered as part of an alarm acknowledgement or even that is was acknowledged. The details of what states are expected to be recovered are provided in the individual object models.