Netbackup 4.5 speed and throughput

[TekieMDX]

We have a sunblade running solaris 9. It is connect to a tape library(timberwolf) that is scsi, and also connect to a CX200 that is fiber. The tape backup and the backup to the disk are approx. the same speed.


Thanks for any help on this situation

 

[Stumpr]

Check out some of the FAQ's in this forum.
This is an especially good one by pgphantom

thread181-27567

Bob Stump
Incorrigible punster -- Do not incorrige

 

[syjl]

Stumpr,

FYI, that thread link does not seem to work right.

TekieMDX, Click on FAQ's at the top of the page and look for the Backup speed FAQ by phantom. It is very helpful.

Next item.

After two days of having a Veritas Architect here, we still can't figure out why we can't get better performance. We get 10mB/sec to tape or disk with NetWare, Solaris, or Windows clients with one stream or many. I understand that there are issues with NetWare and Windows but if I can do a binary ftp at over 30mB/sec, why won't the backups go faster than 10mB/sec. I am trying to troubleshoot this a piece at a time but the network and disks are plenty fast so I don't see the problem. Backups should fly on this equipment.

Here is the test bed

NBU 5.0 MP4
Solaris 8 master (280r) Gigabit network
Solaris 8 media servers (V210's) Gigabit Network. Disk is HP HSG80
Solaris 8 client (280r) Gigabit Network. Disk is HDS 9910

Can someone help me out. Thanks.

 

[mrtimzo]

Hi, we're using NBU4.5 in gig-ether environment. Not sure if you've tuned these parameters, but we've found that performance is indeed modulated with such tweaking..

(1) create /usr/openv/netbackup/NET_BUFFER_SZ text file on your NBU server and all clients, with content thus:

65536

(2) on gig-ether NBU clients, solaris8, kludge network parameters as follows:

/usr/sbin/ndd -set /dev/udp udp_recv_hiwat 65535
/usr/sbin/ndd -set /dev/udp udp_xmit_hiwat 65535
/usr/sbin/ndd -set /dev/tcp tcp_recv_hiwat 65535
/usr/sbin/ndd -set /dev/tcp tcp_xmit_hiwat 65535

I've got a small /etc/rc2.d/S99_ndd_kludge script that does this each time you reboot. Editing /etc/system appropriately may also allow you to do the same thing. Note the syntax will vary slightly based on your solaris platform version.


(3) Clearly if you are using gig-ether you need/want to have jumbo-frames enabled, and of course supported in the switch you are using as well.

General notes from my own experience,

-nature of data being backed up highly affects the backup speed. (many small files vs few large files).

-our config here, we get approx 8-10mb/sec backup of "misc system data files" from solaris clients with 100mbit ether ; 15-20mb/sec backup of "misc fileserver data files" from solaris client with gig-ether/jumbo frames and speed varies on average-size-of-files [approx] ; and best performance we typicaly see is 25-28mb/sec from oracle hot or cold backups, which are streaming large chunks of data (200-400gigs per job typically).

HTH slightly,

--Tim

 

[bignate]

are there performance settings that can be set in the windows servers that are similar to what you just described mrtizmo?

Nate

 

[nbkp25v]

Configure NETBACKUP bptm buffering as per Veritas' recommendation. This topic is discussed in the Netbackup Tuning Guide, technote 240733. The values used here are for IBM LTO Gen I tape drives.

How much shared memory is used by Netbackup?

From the Netbackup Performance Tuning guide:

The maximum amount of shared memory used by Netbackup on a Media Server:
(NUM_DATA_BUFFERS * SIZE_DATA_BUFFERS ) * _of_tapedrives * MPX_for_Stor_Unit

262144 = 32 * 512 * 4 * 4

** Example used to result in numbers listed below **


32 = <install path>\NetBackup\db\config\NUMBER_DATA_BUFFERS
32 = <install path>\NetBackup\db\config\NUMBER_DATA_BUFFERS_RESTORE
262144 = <install path>\NetBackup\db\config\SIZE_DATA_BUFFERS
262144 = <install path>\NetBackup\NET_BUFFER_SZ


Thanks,
Bob