When using trace to get deadlock data the Diagnostic Server creates a trace that is supposed to run continuously. However SQL Server stops the trace every 10 minutes even though we never set any timeout on it. And it does it on a 10 minute boundary like 10:00, 10:10, 10:20 etc. which means it doesn't even run for 10 minutes after it initially starts. So the Diagnostic Server detects that the trace has been stopped and restarts it.
Unfortunately this is logged in the SQL Server error log. The time interval on the deadlock collection has no bearing on the starting or stopping of the trace. The trace runs continuously (except when SQL Server stops it) and stores the deadlock data locally in the collector. The deadlock collection interval merely controls how often the Diagnostic Server asks the collector for the deadlock data.
Disabling the deadlock collection will stop the trace; not immediately but after several minutes. Of course then no deadlocks will be detected.