cancel
Showing results for 
Search instead for 
Did you mean: 

bpbackupdb takes ages

Peter_Jakobs
Level 5
Partner Accredited Certified
Hi,

First I'll explain what I did, I am testing to migrate my master server (tru64Unix) to a new Solaris T2000 box. Solaris 10 Netbackup 5.1mp5

I copied over the /usr/openv/netbackup/db (which is about 44Gb)
In the administration interface I see all the images
In the backup/restore/archive interface I see all the backups and the files being backuped.

If I launch a bpbackupdb it takes more then 24 hours to finish.
If I stop Netbackup and do a tar of the /usr/openv/netbackup/db directorie it finish in 17minutes.

During the bpbackupdb I see that the bpbackupdb process is very busy?
bpdbm and admin logging in verbose 5 doesn't give me any clues.

Any ideas?

Peter
1 ACCEPTED SOLUTION

Accepted Solutions

Stumpr2
Level 6
Compression will make the db smaller and will result in a shorter backup tim. However, you can get a more dramatic change by doing one or both of the following:

DOCUMENTATION: Catalog backups performed using the bpbackupdb command are taking excessive amounts of time
http://support.veritas.com/docs/279908

Details:
Manuals:
VERITAS NetBackup 5.0 Commands for Windows, p. 30
VERITAS NetBackup 5.0 Commands for UNIX, p. 30
VERITAS NetBackup 5.1 Commands for Windows, p.30
VERITAS NetBackup 5.1 Commands for UNIX, p. 27

Modification Type: Supplement

Modification:
Unlike regular backups, which use bpbkar and bptm, the bpbackupdb command does not use any of the buffer tuning parameters. As such, if any tweaks have been used to adjust buffer settings, these performance enhancements will not be used by bpbackupdb.

Two suggestions for working around this bpbackupdb performance limitation:

1. Implement the "Multiple-Tape Catalog Backups" configuration as outlined on page 217 of the NetBackup 5.1 System Administrator Guide for UNIX Volume I (see Related Documents below)
This configuration has the effect of minimizing the amount of catalog data being backed up via the bpbackupdb command. Only the master server's client backups are captured with the bpbackupdb command, and the rest of the catalog is backed up via a policy backup of the master server.

2. Implement the "Catalog Archiving" feature as outlined on page 230 of the NetBackup 5.1 System Administrator's Guide For Unix Volume I (see Related Documents below)
This has the same effect as suggestion 1 (the .f files end up being captured via a user-initiated policy backup) while also drastically reducing the amount of data that bpbackupdb reads from the NetBackup images directory structure.

The disadvantage of either method is that another step is added to the catalog recovery process.

View solution in original post

13 REPLIES 13

Stumpr2
Level 6
Under Solaris operating systems, the catalog backup is very slow.
http://support.veritas.com/docs/273288

DavidParker
Level 6
Bob, do you think Catalog Compression might help this situation out at all?

Just wondering ...

Stumpr2
Level 6
Compression will make the db smaller and will result in a shorter backup tim. However, you can get a more dramatic change by doing one or both of the following:

DOCUMENTATION: Catalog backups performed using the bpbackupdb command are taking excessive amounts of time
http://support.veritas.com/docs/279908

Details:
Manuals:
VERITAS NetBackup 5.0 Commands for Windows, p. 30
VERITAS NetBackup 5.0 Commands for UNIX, p. 30
VERITAS NetBackup 5.1 Commands for Windows, p.30
VERITAS NetBackup 5.1 Commands for UNIX, p. 27

Modification Type: Supplement

Modification:
Unlike regular backups, which use bpbkar and bptm, the bpbackupdb command does not use any of the buffer tuning parameters. As such, if any tweaks have been used to adjust buffer settings, these performance enhancements will not be used by bpbackupdb.

Two suggestions for working around this bpbackupdb performance limitation:

1. Implement the "Multiple-Tape Catalog Backups" configuration as outlined on page 217 of the NetBackup 5.1 System Administrator Guide for UNIX Volume I (see Related Documents below)
This configuration has the effect of minimizing the amount of catalog data being backed up via the bpbackupdb command. Only the master server's client backups are captured with the bpbackupdb command, and the rest of the catalog is backed up via a policy backup of the master server.

2. Implement the "Catalog Archiving" feature as outlined on page 230 of the NetBackup 5.1 System Administrator's Guide For Unix Volume I (see Related Documents below)
This has the same effect as suggestion 1 (the .f files end up being captured via a user-initiated policy backup) while also drastically reducing the amount of data that bpbackupdb reads from the NetBackup images directory structure.

The disadvantage of either method is that another step is added to the catalog recovery process.

Peter_Jakobs
Level 5
Partner Accredited Certified
ndd -get /dev/tcp tcp_xmit_hiwat
49152
ndd -get /dev/tcp tcp_recv_hiwat
49152

Looks like the default on Solaris 10 has changed because I haven't touched these.
I am not using Trunking, but the new Link aggregation:
dladm show-aggr
key: 1 (0x0001) policy: L4 address: 0:14:4f:3c:58:a9 (auto)
device address speed duplex link state
e1000g1 0:14:4f:3c:58:a9 1000 Mbps full up attached
e1000g0 0:14:4f:3c:58:a8 1000 Mbps full up attached

I don't understand what network parameters have to do with the bpbackupdb which can all be done on the local host.

Implement the "Multiple-Tape Catalog Backups" can be a workaround, I even tried it and it's working. But I really think something is wrong because my db is not so big (44Gb) and the machine is really fast.
On other Solaris machines that I have I am doing db backups of 5Gb in about 10 minutes.

Peter

h_m
Level 6
Catalog backups on UNIX are not that slow.....

This migration from Tru64 to Solaris - have you followed all of the steps?

Have you run a normal backup of the netbackup/db directory to see how long it takes?

What other entries do you have in your catalog backup?

Peter_Jakobs
Level 5
Partner Accredited Certified
> Catalog backups on UNIX are not that slow.....
>
> This migration from Tru64 to Solaris - have you
> followed all of the steps?
>

what steps? there is no manual for migrating...
I did a tar of the /usr/openv/netbackup/db and /usr/openv/volmgr/database directory and copied it to the new server.

> Have you run a normal backup of the netbackup/db
> directory to see how long it takes?
>
A normal backup of the db directory takes abou 20 minutes

> What other entries do you have in your catalog backup?
/usr/openv/netbackup/db
/usr/openv/volmgr/database
/usr/openv/var

h_m
Level 6
Can you run a manual backup of the catalog via the catalog/manual backup option and see how long it takes?

h_m
Level 6
Have you also moved /usr/openv/volmgr/database ?

Peter_Jakobs
Level 5
Partner Accredited Certified
A manual backup takes more then 24 hours.

Yes I also moved the volmgr database.
It's the database with all the tapes, so I have to copy them as well.

h_m
Level 6
Put a call into Veritas - that is not right at all...

h_m
Level 6
I'd also make sure you're copying over the right files from one flavour of UNIX over to another flavour...

Peter_Jakobs
Level 5
Partner Accredited Certified
After sending lots of verbose, explorer and dtruss output to the Sun support center, they told my to set the :

set ip:do_tcp_fusion=0
in the /etc/system file.

With this parameter the backup is taking about one hour (45Gb), so it is acceptable (not very fast though).

Thanks to everybody who tried to help me.

Peter

Stumpr2
Level 6
Peter,
Thank you for posting a follow-up. It helps everybody in the forum when a follow-up gets posted. Thanks again.
Bob