For now, the best you could hope for would be an image mode snapshot backup
going into a deduplicated storage pool.
Alternatively, if your application has a TSM connect agent (such as TDP for
Databases), then you could use that to get the application data incrementally
rather than pulling the blocks.
Otherwise, if you have a home-made application, it shouldn't be hard to write
an incremental image backup tool.
Use OS and/or Array commands to:
* Quiesce the filesystem
* Lock writes
* snapshot
* resume r/w
* present the snapshot devices
Then, using the code from adsmpipe, hash procedures, and standard file I/O:
* Try to pull an existing hash table out of TSM. Backup 0 would have none.
* Tead in the snapshot raw logical volume and generate hashes for every block.
* For any block without a hash, write out in binary format the sector offset,
number of blocks, and then the raw data
* Once the whole "image" is written, then you can save the hash table as a
second file.
* If you wanted to be fancy, you could index by hash and gain some manner of
block-level dedupe.
Finally, you'd destroy the filesystem snapshot if necessary.
Hash code and file I/O code are readily available on the internet. adsmpipe
can also be found on the internet.
I would recommend static linking if your UNIX of choice prefers it. On Linux,
ADSMPIPE is very sensitive to libc changes.
As for something like FastBack Mounter, you'd need to write a block level
device driver, which is more complicated.
________________________________
From: Mehdi Salehi <iranian.aix.support AT GMAIL DOT COM>
To: ADSM-L AT VM.MARIST DOT EDU
Sent: Sat, January 23, 2010 1:34:40 AM
Subject: [ADSM-L] sector-based incremental backup of filesystem
Hi,
Is there any way for sector-based incremental image backup in Unix systems?
B/A client helps incremental-by-date, but it is inefficient for some
applications. What I am looking for is a feature like what FastBack or
UltraBac present for Windows.
Thanks
|