Eric,

If you look in your bptm logs it will display the exact MB/sec at the end
of every job..

you should also look in the bptm for these messages at the end of each
backup:

12:55:42 [4104] <2> write_data: waited for full buffer 0 times, delayed 0
times

this will tell you whether or not the tape drive is waiting on data from
the disk.  ie, if it says "waited for full buffer 20000 times" that would
indicate that there is a bottleneck on getting the data off of the disk.

If it states that it "delayed <a bunch of> times" that that would indicate
that your tape buffers are filling up and netbackup is having to delay
reading data off of disk because it is waiting on a free buffer.  Your
bottleneck is somewhere on the tape side, tweaking your tape buffers (see
below) should help this unless your SCSI interface is the bottleneck.

I would use sar or iostat while the backups are running to see whether or
not the disk's are running at 100% utilization.

I would also check the CPU's utilization as well to make sure they aren't
being taxed..  (using iostat or sar again)

If you can find that those are not the problem, then you may need to
increase the buffers going to your tape..  these can be set by creating
these two files, each with a single number as it's only content:

/usr/openv/netbackup/db/config/SIZE_DATA_BUFFERS
/usr/openv/netbackup/db/config/NUMBER_DATA_BUFFERS

SIZE_DATA_BUFFERS is the size of each of the buffers writing to tape.
this is defaulted at 32k.  A decent number to change this to would be 64k
or something larger.  Note, that this will also change the block size of
the data written to tape.  Definitely test out your restores after you
make a change to this value to ensure that things still work..  The value
to place in the file is in bytes, so to set it at 64k you would place the
number 65536 in the file.

NUMBER_DATA_BUFFERS is the number of these buffers to have for each
backup.  default is 8 (I think)..  you may want to increase this to 32 or
so.. I would increase it in small increments and watch to see if the
performance is increases at all.

NOTE:  that changing these values changes the amount of shared memory you
will be allocating for each backup.

if you have 32 buffers at 64k each, EVERY backup will use 2MB of memory
(64k*32). 

Also, be sure that you are writing to your tapes with variable-length
writes when modifying the tape buffer size (on a sun solaris system, this
is set in your /kernel/drv/st.conf file)

Hope that helps,

Jon

On Wed, 1 Nov 2000, Porter, Eric wrote:

> Rob,
> 
> Answers to your questions are below.  
> Your comments/suggestions (or anyone else's) are appreciated!
> 
> Thanks
> 
> Eric
> 
> **** Where is the data coming from as it's being written to these 9840's?  
> 
> The data is being backed-up out of raw logical volumes.  These volumes are
> on the same server as netbackup  (i.e. local) (They are copies generated
> using emc timefinder).
> 
> **** How many 9840's are you using for backups?
> 
> I have 8 drives, I am using 7.  The scripts I am using split the list of raw
> logical volumes into 7.  Then we start 7 backups each to tackle one set of
> raw logical volumes.  The 8'th drive is reserved for backups of oracle
> archive logs (done hourly).
> 
> **** Where is your data coming from?  Is it local on this media server, is
> it on a SAN, is it on one or more segments of your LAN, etc?  
> 
> Local for the important data  (about 2 Terabytes).  
> Some less serious stuff originates on a dedicated switched 100M Ethernet
> (running full duplex).
> 
> **** Are you configured to use Multiplexing?
> 
> No... I am pushing a raw partition from the same system I figured
> multiplexing would hurt not help.
> 
> **** with a 9840 drive (transfer rate of 10-20 MB/sec depending on
> compression AND STREAMING) it would be real easy to 'underfeed' your drives
> and therefore cripple your performance.
> 
> If the numbers in the netbackup GUI are to be believed I am getting 4 -- 5
> MB/sec per tape drive.
> _______________________________________________
> Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
> http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
>