You have a disk array copy of the data, is that located close or far?  Have you 
considered a disk array snap shot also?
If you perform a journaled file system backup and an image backup then you 
should be able to restore the image and then update the image with the file 
system restore.  This might take a long time, I have never tried it.
What failure are you trying to protect against?  In our case we use the disk 
arrays to protect against a data center loss and a corrupt file system and a 
TSM file system backup to protect against the loss of a file.  Our big ones are 
in the 10 million file range.  Using a 64bit Windows server we can backup the 
file system in about 6 - 8 hours without journaling.  We suspect we could get 
the time down to around 4 hours if the TSM server was not busy backing up 500 
other nodes.

To me the important thing is to figure out what you are protecting against with 
each thing you do.  Also be sure and ask what the Recovery Point Objective 
(RPO) is.  If it is less than 24 hours then array based solutions may be the 
best choice.  Over 24 hours then TSM may be the best choice.

Andy Huebner

-----Original Message-----
From: ADSM: Dist Stor Manager [mailto:ADSM-L AT VM.MARIST DOT EDU] On Behalf Of 
Mehdi Salehi
Sent: Thursday, April 30, 2009 9:39 AM
To: ADSM-L AT VM.MARIST DOT EDU
Subject: Re: [ADSM-L] best backup method for millions of small files?

Hi,
None the two methods that you mean in the user's guide are suitable for my
case. "Image+normal incremental" that you emphasized in your post means
getting full image backups for example every week. For the incremental part,
one file-based full backup is needed which is a nightmare for 20 millions.
OK, if I accept the initial incremental backup time (that might take for
days), what happens in restoration?

Naturally, last image backup should be restored first and it will take A
minutes. Provided that image backups are weekly, the progressive incremental
backups of the week is about 6*20MB=120MB. Now imagine 120MB of 15-20K files
are to be restored in filesystem with an incredibly big file address table
and system should create an inode-like entry for each. If this step takes B
minutes, the total restoration time would be A+B. (A+B/A) ratio is important
and I will try to measure and share it with the group.

Steven, your solution is excellent for ordinary filesystems with a limited
number of files. But I think for millions of files, only backup/restore
method that do not care how many files exist in the volume are feasible.
Somehing like pure image backup (like Acronis image incremental backup)  or
the method that FastBack exploites.

Your points are welcomed.

Regards,
Mehdi Salehi


This e-mail (including any attachments) is confidential and may be legally 
privileged. If you are not an intended recipient or an authorized 
representative of an intended recipient, you are prohibited from using, copying 
or distributing the information in this e-mail or its attachments. If you have 
received this e-mail in error, please notify the sender immediately by return 
e-mail and delete all copies of this message and any attachments.
Thank you.