There are 3 "service requests": rename, delete, and a check to see if
a file exists.  The ckpt server normally gets lots of delete "service
requests" returning DOES_NOT_EXIST because the schedd and shadow are
extra paranoid about cleaning up checkpoint files for
removed/completed jobs.  (They aren't paranoid enough, since the
checkpoint servers still fill up with old checkpoints after a while.)
You should only get nervous if a "restore request" fails with
DOES_NOT_EXIST.  So, it's normal to see lots of entries that look like
this:

  9/22 10:44:55   Receiving service request from 128.105.146.14
                  Using descriptor 5 to handle request
  9/22 10:44:55   Service request cannot be completed:
                  Attempt returns error #91

There's no good way to clean up the checkpoint server disk.  The least
bad way we've found is to delete checkpoint files that are old.  In
the checkpoint server directory, do:

  find *.*.*.*/* -atime +90 -exec ls -l {} \;

Look at the resulting checkpoint files carefully.  Once you're
convinced it's safe to proceed, cross your fingers and do:

  find *.*.*.*/* -atime +90 -exec ls -l {} \; -exec rm {} \;

This example deletes files more than 90 days old.  Feel free to
replace 90 with a more appropriate age.
