> Does anyone have information on tuning the timeouts for 
> Windows Incremental backups? My environment has Windows 2003 
> clients, running NetBackup 6.0 MP5, sending data to a Windows 
> 2003 64bit Media Server, running Netbackup 6.0 MP5. 
> Some of the Incremental backups are hanging and some report 
> the "couldn't write KEEPALIVE to COMM_SOCK" error.

Everything below presupposes that the failures are "too much directory,
too few changed files" and that lengthening keepalives is the answer.

On Solaris, the -get retrieves the current value and the -set changes it
(times in milliseconds, I believe.)  

# ndd -get /dev/tcp tcp_keepalive_interval
7200000
# ndd -set /dev/tcp tcp_keepalive_interval 1800000

I'm sure there are archives in veritas-bu for this, plus technotes on
the support site (look for "keepalive").

MS article on modifying a different tcp parameter; might be a good
start:    http://support.microsoft.com/default.aspx?scid=kb;en-us;170359

In my experience, a large and complex NTFS system is terrible at
traversing the directory tree, and as you surmised, you can timeout
before finding the next changed file.  A kludge I used for one Windows
box where changing TCP parameters wasn't an issue but duration of
incremental backups was, is to run a script that does "cd \ & dir /s
>nul" on the problem drive(s) since the filesystem seems to cache
directory information well.  (As a point of information, I've had that
script run 42 minutes the first time and less than a minute--I
forget--the second time.)


_______________________________________________
Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu