Yes media servers.

The vendor published speed is usually conservative.
For example, sometimes I get 24MB/sec to 9840FC's which
have a native throughput of 10MB/sec.

Therefore, I calculate buffers based on the theoretical as
opposed to the conservative figure.

My understanding is, if backups were written at 64k
they will be read at 64k. However, with 256k buffers
you will have more buffer space free. In addition, Netbackup
should start writing backups at 256k blocks.

You have to be careful though, because my understanding is
netbackup will use the block size setting to "read from disk"
as well. In which case 256k may be optimum for tape but not
so for your disk configuration. Im not aware of a setting that
allows you to set them differently.

There is another buffer setting I havent seen documented
called NUMBER_DATA_BUFFERS_RESTORE. The name says it all.
However, Im not sure if restores reference this file or
NUMBER_DATA_BUFFERS. It would be straight forward to work
out. Simply kick off a restore and check the size of
the shared memory segment allocated to netbackup using
"ipcs -a". Divide the size by the block size and number
of tape drives used to satisfy the restore, and you should get
the number of data buffers uses. Then see if this figure is
equal to NUMBER_DATA_BUFFER or whether it is the default 16.
If its the default 16, then create NUMBER_DATA_BUFFERS_RESTORE
and test again.

Regards
Peter Marelas

-----Original Message-----
From: Jason Ahrens [ahrensj AT psi DOT ca] [mailto:AhrensJ AT psi DOT ca]
Sent: Wednesday, 9 May 2001 3:09 AM
To: Veritas-bu AT mailman.eng.auburn DOT edu
Subject: RE: [Veritas-bu] NBU Data Buffer sizes


I am assuming this is on the media server? Or does it matter? Or should it
be on both? (assuming separates master/media servers)

As a check on my own math, I have StorageTek 9800 drives. 10MB/s theoretical
maximum write speed, 256k block size.

At some point in the past, I set
/usr/openv/netbackup/db/config/SIZE_DATA_BUFFERS to contain 262144 (256k)...

I did this on the master server. I expected restores to break for the 'cut
over' time after I increased blocksize from 64k to 256k, but it did not...
(should it have? Does this mean I'm still using a 64k block size?)

Now, using the below system, I should set
/usr/openv/netbackup/db/config/NUMBER_DATA_BUFFERS to be
(10MB/256k)=(10485760/262144) which would give me 40 buffers (per drive)...


Jason

> -----Original Message-----
> From: Marelas, Peter [mailto:MarelP AT AUSTRALIA.Stortek DOT com]
> Sent: May 7, 2001 20:05
> To: 'Tim Allen'; Veritas-bu AT mailman.eng.auburn DOT edu
> Subject: RE: [Veritas-bu] NBU Data Buffer sizes
> 
> 
> Tim,
> 
> I tend to tune the buffers based on the maximum theoretical throughput
> of the tape drive in question.
> 
> However to do so you need memory on the server.
> 
> You've got it set to 16 and 64512 respectively.
> 
> Therefore Netbackup will use 16*64512=1032192 bytes per tape drive
> in use.
> 
> Assuming the maximum theoretical throughput of a drive is 6MB
> I would set SIZE_DATA_BUFFERS to the tape drives optimum block
> size (usually 64k, 128k or 256k).
> 
> Lets assume its 64k (65536), divide 6291456 bytes (6MB) by 
> 65536 bytes (64k)
> you
> get 96 NUMBER_DATA_BUFFERS. That should be the optimum setting.
_______________________________________________
Veritas-bu maillist  -  Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu