I had a situation where an Exchange server completely stopped responding even though SMTP, POP3, etc. appeared to all be functional. Of course, no email was being received and no users were able to check messages through any Exchange interface. When I looked at the event log, MSExchangeDSAccess had logged EventID 2103:
Process MAD.EXE (PID=1111). All Global Catalog Servers in use are not responding:
Doing some research led me to determine that this was a Kerberos error. On a hunch, I checked the clocks of the two systems (the Exchange server is also a domain controller) and determined that they had skewed to the point that each was rejecting the other. I’m unsure how they got out of sync, I suspect one did an NTP sync and the other didn’t pick up on it. I set the clock of the lagging server to be the same minute as the other one, rebooted, and presto, everything came back up.