Re: [Veritas-bu] Performance tuning with remote NDMP.
2009-06-17 18:07:59
william.d.brown AT gsk DOT com wrote:
Well NDMP logging is done differently, so you may want to search for the
technotes for that - it will likely give more information. However, I've
heard that it can produce an enormous amount of logging.
I know. I've already filled up the file system once :-)
I've not tried remote NDMP any time recently, so I can't claim real-world
experience. I have read the NetApp documents, and you'll find them on the
filer if installed. In particular there is information about the
parameters that can be set by adding the SET commands to the backup
selection list. Really there is very little, but I do recall it saying
that for NDMP you could not change the block size from 64KB even though
the setting exists. Now I'm not sure how that applies to remote NDMP, but
I'm sure you'll either see that in the logs. I suspect that somewhere in
the path, even though you are using the tape mover in your Media Server
and the disk mover in the NetApp, that 64KB will be in force.
That is going to limit the bandwidth and slow you down.
I'll dig up the NDMP manual and give it another look. From what I saw
it seemed to treat remote NDMP the same as any other client backup with
a few differences.
You are also putting over the same link I guess the backup data and the
catalog information which AFAIR is sent at each change of path to the
'NetBackup for NDMP' server, in this case your Media Server. That has to
grab the information and send it on to the Master Server each time, so the
catalog gets updated. I understand that every time it does that there is
another DNS lookup - as you turn up the logging you will see a lot of
activity. This is one reason why it is recommended not to put the NDMP
agent on the Master.
The master server is the media server so I wouldn't think this would
hurt too much.
You might look at the network setting on the RHEL server to make sure
you've done all the TCP tuning of receive buffers and the like. You
might be able to turn on jumbo frames on the filer and the RHEL server, if
your network allows. I'd say 110MB/s was pretty good, wish I got that
anywhere.
Unfortunately the jumbo frames are not an option at this point. I have
looked into doing a direct connect between the NetApp port and the
master serve 10G ports which would allow for the jumbo packets. That
presents other logistical issues that I can really work around right
now. The 110 MB/s isn't bad until I put it in perspective of having to
backup 70+ TB of data with a number of 6 TB file systems mixed with a
bunch of smaller ones. No multiplexing of NDMP streams means I'm going
to need a lot of drives to get things to run in a timely manner.
Thanks,
Jeff
William D L Brown
veritas-bu-bounces AT mailman.eng.auburn DOT edu wrote on 17/06/2009 20:58:29:
Greetings,
I found some posts from April about people trying to get NDMP
performance out of a NetApp but seem to stall out around 120 MB/s.
I didn't find any posts that detailed why. I know that having a
trunk connection of 1 gig ports will still only use 1 gig port for
the point to point connection and not use the entire aggregated
bandwidth. I didn't know if this was the issue or not.
Our setup is a 6030 running 7.2.5.1 Ontap and a linux RHEL4 server
running 6.5.3. Both the NetApp and the master server had 10G
network cards in a private network. When I do a dump to null on the
filer I can average over 200 MB/s. When I do a dd to a tape drive
on the master I can get over 120 MB/s to a Gen 3 LTO. When I do
backups of a single file system I can never exceed 70 MB/s. Running
multiple backup jobs has maxed out at 110 MB/s. I have verified
through the network switch traffic that the data is going between
the 10G ports.
I've gone through the performance tuning guide and played with the
buffer settings for NET_BUFFER_SZ, SIZE_DATA_BUFFERS_NDMP,
NUMBER_DATA_BUFFERS, and SIZE_DATA_BUFFERS. I can make performance
worse but can never get better than what is listed above. One thing
I've noticed is that even though I've made changes to the VERBOSE
statement in the bp.conf file and restarted NetBackup, I never get
the wait and delay messages in the bptm logs. The master server
properties have all be set to maximum logging Obviously there is no
bpbkar log file on the filer to look at. Am I missing something to
get that turned on?
Thanks,
Jeff
--
Jeff Cleverley
Unix Systems Administrator
4380 Ziegler Road
Fort Collins, Colorado 80525
970-288-4611
jeff.cleverley AT avagotech DOT com
_______________________________________________
Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
-----------------------------------------------------------
This e-mail was sent by GlaxoSmithKline Services Unlimited
(registered in England and Wales No. 1047315), which is a
member of the GlaxoSmithKline group of companies. The
registered address of GlaxoSmithKline Services Unlimited
is 980 Great West Road, Brentford, Middlesex TW8 9GS.
-----------------------------------------------------------
_______________________________________________
Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
--
Jeff Cleverley
Unix Systems Administrator
4380 Ziegler Road
Fort Collins, Colorado 80525
970-288-4611
jeff.cleverley AT avagotech DOT com
|
_______________________________________________
Veritas-bu maillist - Veritas-bu AT mailman.eng.auburn DOT edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu
|
|
|