We have several dozen SQL Servers, with multiple databases per server, that
are backed up nightly using NetBackup 6.0 MP4. Periodically, we will have
individual database backups fail with a status 42. The detailed status for
the job will show the following:

6/24/2008 10:20:13 PM - started process bpbrm (2748)
6/24/2008 10:20:13 PM - connecting
6/24/2008 10:21:01 PM - started process bpdm (6620)
6/24/2008 10:21:01 PM - connected; connect time: 00:00:48
6/24/2008 10:21:09 PM - started process bpdm (6632)
6/24/2008 10:21:09 PM - begin writing
6/24/2008 10:21:40 PM - Error bpdm(pid=6632) socket operation failed - 10054
(at child.c.1172)
6/24/2008 10:21:40 PM - Error bpdm(pid=6632) unable to perform read from
client socket, connection may have been broken
6/24/2008 10:21:41 PM - end writing; write time: 00:00:32
network read failed(42)

All other databases on the server will back up successfully. If we rerun
the backups, the 'failed' database will generally back up without issue.

We see this error probably 5 times a week. It appears on several different
server, with several different databases, and there does not seem to be any
discernible pattern. There are no network errors in the Windows system logs.
In fact, if we look at the SQL Server logs, they will show that the backup
completed successfully:

Database differential changes were backed up. Database: XXXXX, creation date(time):
2008/05/20(10:14:05), pages dumped: 128, first LSN: 69:356:34, last LSN:
69:371:1, full backup LSN: 69:335:37, number of dump devices: 1, device information:
(FILE=1, TYPE=VIRTUAL_DEVICE: {'VNBU0-692-1216-1214284317'}). This is an
informational message. No user action is required.

Can someone point me to a possible course of action to resolve these issues?