Veritas-bu

[Veritas-bu] Netbackup benchmark & performance tuning

2007-02-28 07:40:04
Subject: [Veritas-bu] Netbackup benchmark & performance tuning
From: Ueli.Schweizer at AGITE-Software.com (Ueli Schweizer)
Date: Wed, 28 Feb 2007 13:40:04 +0100
Hi Vijay

"Waited for full buffers" messages during a backup means that the tape is
able to write more data than bpbkar is able to send. In this example the
"11254 delays" indicates, that the backup would finish 11254 * 30 msec = 338
seconds faster under <<optimum>> conditions. Of course it is very important
to know how much data you've backed up during this attempt to find out if
this number is extremely high or not. But anyway: it indicates that the
filesystem can not be read as fast as the data can be written to tape.

A question to your performance test using bpbkar: What path are you using as
<filesystem>? If you are using the device path (e.g. /dev/dsk/c1t0d0s5) you
are reading the raw disk. Only if your using the mountpoint (e.g. /usr) you
are reading the filesystem. Depending on the fragmentation of a disk the
differences in performance are just huge. For me the number you measured
(170 MB/s) seems to be very (actually VERY) high for filesystem backup, so
probably you're backing up raw disk data or the measured performance was
affected by read cache or somewhat else.

For me the 22 MB/s looks much more realistic than the 170 MB/s. For FS
backup in general and non-RAID OS disk in special I think 22 MB/s is just
average and I'm not sure it there is much tuning on this. I would focus on a
tuning configuration parameters including MPX.

Cheers

Ueli Schweizer

AGITE Software AG . Boesch 43 . CH-6331 Huenenberg . Switzerland
Direct: +41 79 204 9190 . Phone: +41 41 781 5678 . Fax: +41 41 781 5677
Ueli.Schweizer at AGITE-Software.com . http://www.AGITE-Software.com 

backupVISUAL ...
                  ... we make life on backup easy
-----Original Message-----
From: veritas-bu-bounces at mailman.eng.auburn.edu
[mailto:veritas-bu-bounces at mailman.eng.auburn.edu] On Behalf Of Vijay Kumar
Sent: Wednesday, February 28, 2007 12:55 PM
To: bob944 at attglobal.net; veritas-bu at mailman.eng.auburn.edu
Subject: Re: [Veritas-bu] Netbackup benchmark & performance tuning

Bob, Paul, murtuja

Some clarification. Backup of local server means
Netbackup Master/Media server itself not any other
system from the network. 

I am trying to backup OS Disk. NO raid configuration.

# df -kFufs
Filesystem            kbytes    used   avail capacity 
Mounted on
/dev/dsk/c1t0d0s0    8263373  180817 7999923     3%   
/
/dev/dsk/c1t0d0s5    19623529 1458826 17968468     8% 
  /usr
/dev/dsk/c1t0d0s3    10327372 7534947 2689152    74%  
 /var
/dev/dsk/c1t0d0s4    15493995   30262 15308794     1% 
  /opt
/dev/dsk/c1t1d0s0    41311843 5549567 35349158    14% 
  /usr/openv


Following error message i am getting in bptm logs

11:54:09.851 [927] <2> write_data: waited for full
buffer 7898 times, delayed 11254 times

May be this is the bottleneck. So i tried to configure
different buffer size but still messages.

I checked the Disk I/O by using 
/usr/openv/netbackup/bin/bpbkar -nocont -dt 0
-nofileinfo -nokeepalives filesystem > /dev/null
Here I am getting Disk speed more that 170Mbps

Yes Bob its in magabytes/second

Thanks & regards,
Vijay


--- bob944 <bob944 at attglobal.net> wrote:

> First, congratulations on supplying a lot of useful
> info in the first
> posting.
> 
> > We are using Netbackup 5.1MP6 on Solaris 9.
> > H/W Detasils : 
> > Master/Media Server : v440, 4CPU, 16Gb RAM
> > Tape Library : L500, 4 F-CAL Drive (LTO Gen 2)
> > Network Speed : 1GBPS
> > Tape library is directly connected to server.
> > 
> > Currently when trying to take backup of local
> server i
> > am getting 22Mbps throughput which i feel is very
> > less. 
> 
> Not to quibble, but I'm sure you are saying 22
> megabytes, not megabits,
> right?  (I know someone who sized his tape HBA
> requirements thinking the
> specs were MB/s, when they were, of course, Mb/s--so
> it's worth being
> rigorous on the notation.)
> 
> > Current Buffer settings:
> > NET_BUFFER_SZ : 262144
> > NUMBER_DATA_BUFFERS :128
> > SIZE_DATA_BUFFERS : 262144
> > 
> > /etc/system:
> > *Added for Nebackup
> > set msgsys:msginfo_msgmap=512
> > [...]
> > 
> > So i would like to know  practically what is the
> > maximum throughput i can get & buffer setting.
> What
> > all changes i need to do on server to get maximum
> > throughput
> 
> LTO2 drives aren't all that fast--half-height LTO2s
> are
> manufacturer-rated at something like 24-28MB/s
> native and the best
> rating I know of is 35MB/s.  See point #2 below.
> 
> You want three things:  
> 1.  NetBackup 6.0 Performance Tuning Guide
> 2.  Follow the guide to determine where the
> bottlenecks are--there's no
> point in tuning NetBackup buffers if, say, the
> client can't supply data
> fast enough.
> 3.  When you can demonstrate to yourself that
> NetBackup is the slow link
> in the chain, then it is worth a controlled,
> rigorous buffer tuning
> exercise.
> 
> 128 256KB buffers is way more than enough to drive
> an LTO2.  
> 
> 



 
____________________________________________________________________________
________
Don't pick lemons.
See all the new 2007 cars at Yahoo! Autos.
http://autos.yahoo.com/new_cars.html 
_______________________________________________
Veritas-bu maillist  -  Veritas-bu at mailman.eng.auburn.edu
http://mailman.eng.auburn.edu/mailman/listinfo/veritas-bu