Veritas-bu

[Veritas-bu] NetBackup tuning question

2003-08-30 14:08:08
Subject: [Veritas-bu] NetBackup tuning question
From: larry.kingery AT veritas DOT com (Larry Kingery)
Date: Sat, 30 Aug 2003 14:08:08 -0400 (EDT)
> I started this thread a while ago and got sidetracked.  I am running
> NetBackup FP 4.5 on HP-UX 11.i, master server is on an HP L2000, and
> this media server is on an HP N4000 with a SureStore 20/700 library
> using LTO-1 drives.  The majority of my backups are coming off split
> BCV's mounted to my media server from an EMC unit.  After a bit of
> digging and analysis we have come to the conclusion that we need to
> do a bit of tuning.  We are only getting around 3 to 5 mb/sec on
> backups according to the Activity Monitor gui.  We ran a test
> outside of NetBackup and pumped around 26 mb/sec compressed data and
> 14 mb/sec uncompressed to one of the drives.  This kind of points a
> finger at NetBackup.
>
> First of all, is there a NetBackup Performance Tuning guide that
> refers to NetBackup FP 4.5?  All I can find is the one for Version
> 3.2/3.4.  My biggest fear is that the tuning guide it has statements
> about possible restore problems after changes to:
> NET_BUFFER_SZ
> NUMBER_DATA_BUFFERS
> SIZE_DATA_BUFFERS
> They only say there may be a problem, well does this mean some of my
> restores may work and some may not?  I do plan to test restores

The issue is when you increase SIZE_DATA_BUFFERS greater than what the
OS or tape drive can handle.  Usually what will happen is that the OS
will silently break up those large blocks into smaller blocks.  Now,
come restore time when NBU positions to block X, it's going to end up
in the wrong place because it doesn't know that the OS actually wrote
like X*4 blocks or something.

I think newer (4.5 and probably 3.4_4) will actually catch this at the
end of the fragment and freeze the tape, but I haven't tested this
myself.  

> after any changes but there is no way to test a restore for everyone
> of my policies.  The guide also has the following warning: "NOTE:
> Restores use the same buffer size that was used to backup the image
> being restored."  Is this saying that to do any restore I will have
> to change the above parameters back to what they were when that
> backup was taken or will NB determine the backup size and use that
> for the restore?  We have been running NetBackup for a while now and

SIZE_DATA_BUFFERS is ignored at restore.  Whatever size was written to
the tape is used instead.

> I don't want to jeopardize my ability to do restores when needed.
> If anyone has any experience with changing the above, please share
> what you know.  Thanks in advance.

Before you start changing these, you first need to determine where the
bottleneck is (disk/tape/etc).  There was a comment on here a year or
so ago about serious perf differences when using character device vs
block device when backing up raw partitions, but I can't seem to find
it right now.

-- 
Larry Kingery 
                         This space for rent

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