We do what Mark does. Scheduled job to stop the service, give it plenty of time to do the backup (run a test first and add a couple of hours), run whatever backup (we use windows backup), scheduled job to restart the service. Over the last couple of years I remember it not working twice.