NDMP Backups

[Project_Man]

ok we have approx 4TB (approx 23 million files) of data on our NDMP device.

We currently have numerous polocies setup for NDMP but have recently found that we are limited to 4 streams.

We used to just have 4 poloicies but they were taking 18-24hours to complete for incrementals

Each policy has the following:

NEW_STREAM

Set Hist=y

Set Snapsure=y

Now I see there is a

Set type=tar &

Set type=vbb

What do these options do and what advantages/disadvantages do they give?

Would I be better off with s smaller number of policies?

[AAlmroth]

If the filer is an EMC Celerra, then the flags have following meaning;

HIST: File history, required far fast restores, aka Direct Acces Restore (DAR).

You seem to miss the SET DIRECT=y line though.

SNAPSURE: A snapshot is created prior the actual backup (good for consistent backups running for many hours).

TYPE: tar, dump or vbb; the type of backup, tar is file based, whereas VBB is volume block-level backup. VBB is generally faster.

You should also add OPTION=nt if it is a file system for CIFS shares.

Also, if you have a backend Centera, you should not backup archived items, so use SET EMC_OFFLINE_DATA=n. Backups usually slow down to ~2MB/s if having to retrieve archived files.

If you go the the EMC site, you will find a lot more information about the various options available.

/A

[Project_Man]

What is SET DIRECT=y used for?

i assume if vbb is available and it is as I using Celerra you would use it over tar and dump?

Is there any disadvantages to vbb?

[MortenSeeberg]

Just fell on this post; Andreas is a busy guy :) you might not care any more, but just to keep the history:

set DIRECT=Y ensures that DAR is enabled (ability for the tape to wind directly to where the file is on the tape)

VBB is a block-level backup, but IF you have deduplication enabled on the volumes you are running VBB backup then you cannot do single-file restore, as the data you are backing up are deduplicated. Therefore the recommendation is still dump/tar for deduplicated volumes on Celerra.

If you are using the Celerra for CIFS shares, you might also be interested in saving the ACL info:
set OPTION=NT

Regarding having several policies that is always a tradeoff, fewer policies are easier to administer, but also takes longer to complete, and especially if you are having problems with failing backup or to long running backups then you split.

 

[Deepak_W]

This is very common problem for most of the NDMP backup scenarios. Higher the number of files in one NAS volume higher the time required for NetBackup to collect the metadata. This affects backup window in big way.

This is why we suggest our customers to split the NAS volumes in small small volumes. This does not affect their user accessability as users are concerned about their share / export path, which NAS admin can easily emulate to keep the same as previous.

This also enables NBU to process these volumes parallelly hence completing the metadata processing in less time as compared to having single NAS volume.

Give a thought on this. Hope this helps.

[Jacob]

It can take a very long time to restore from a backup taken with vbb if you have high fragmentation on your volumes