Hi Folks,
Thanks so much for all of your suggestions regarding our 6040 outage. As always, very helpful.
While this may not explain the outage, we have noticed some I/O load spikes coming from our email servers since the outage. We have tracked it down to very inefficient processing of mailboxes. Apparently one mail client (Outlook) is deleting multiple messages from a mail folder with a sequence of individual message delete commands. Instead it could use a single command with a list of messages to delete. The mail server is Communigate (not Exchange). The mail folders are flat files so whenever messages are deleted, the tail end of the folder gets rewritten, minus the deleted messages.
Yesterday a single 1.5 GB folder was being read from the filer and written back to the filer again and again to delete numerous messages near the front, one at a time. The user confirmed that she had marked a bunch of messages for delete and then committed the change. We got some complaints about slow email performance from folks who were connected to that particular email server (we have a cluster). But everything else was OK, and the filer handled it gracefully.
It would be nice if Communigate would logically delete messages and rewrite folders in a background process after some number of logical deletes have accumulated. Maybe that is a config option...
Steve Losen scl@virginia.edu phone: 434-924-0640
University of Virginia ITC Unix Support