Veritas-bu

[Veritas-bu] buffer settings for bpduplicate?

2002-10-10 10:36:16
Subject: [Veritas-bu] buffer settings for bpduplicate?
From: jaearick AT colby DOT edu (Jeff A. Earickson)
Date: Thu, 10 Oct 2002 10:36:16 -0400 (EDT)
Hi,

   I have a Sun L20 tape robot with two DLT8000 tape drives, connected
to a Netra 105 running Solaris 8, with NB 4.5.  I have tuned my Netbackup
buffer settings per Veritas technote 183702 to be NUMBER_DATA_BUFFERS=16 and
SIZE_DATA_BUFFERS=65536.  When I took the NB Advanced Course, they talked
about "waited for full buffer" and "waited for empty buffer" messages in
the bptm logs, and how "waited for full buffer" messages are really, really
bad -- because the tape drives will stop a lot and "shoeshine".

   The NUMBER_DATA_BUFFERS and SIZE_DATA_BUFFERS do a good job of minimizing
"waited for full buffer" messages for ordinary backups, but I still get
a ton of waits for bpduplicate jobs.  I would guess that that the size
and/or number of buffers for a tape-to-tape copy within the robot are
still not big enough to give me good performance.  The command I use
for doing duplicates is:

bpduplicate -v -dstunit slate-dlt-robot-tld-0 -dp duplicates \
        -hoursago 11 -L /tmp/dup2.output

and the originals are not multiplexed.  I've read the manpage for bpduplicate
and don't see any option to specify the buffers settings for this job.

   Anybody have any good ideas as to what the buffer settings should be
for a high-performance duplicate between two DLT8000's?  Any ideas on
how to reset the buffer settings for bpduplicate, short of script
manipulations before the bpduplicate job runs?

** Jeff A. Earickson, Ph.D                         PHONE: 207-872-3659
** Senior UNIX Sysadmin, Information Technology    EMAIL: jaearick AT colby DOT 
edu
** Colby College, 4214 Mayflower Hill,               FAX: 207-872-3076
** Waterville ME, 04901-8842
----------------------------------------------------------------------------


<Prev in Thread] Current Thread [Next in Thread>