My experience matches your understanding.

My limited experience with large large backups has led me to avoid them 
where possible!

In case you think I'm an idiot, you're not wrong, but maybe the advice 
can be helpful to a few ...

I'll give 3 examples from recent activities in my own shop:

  * I had a mail server with 250+GB in 15+M files.  It took 60+ hours 
for a full backup.  It turns out that its data is spread over many file 
systems.  I turned on "allow multiple streams" and the total backup time 
was reduced to about 12 hours (It will be substantially better than that 
if I'm able to split two of the larger file systems).

  * I have a development system that had 300+GB of data in dozens of 
Oracle DBs.  Full and incremental backups are similar in size and very 
long in duration.  Many of the DBs are also backed up with the NetBackup 
for Oracle agent.   Backups are being brought under control by "allow 
multiple streams", as well as excluding the Oracle DB files from file 
system backups.

  * Less successful ...  with one large file server (large because it's 
currently at the end of a relatively slow pipe), I've carved up the file 
system using multiple policies and exclude files so each policy exists 
for a part of the data (make sure one policy is "all the rest" or you'll 
get killed by new data/structure).  This works, but is, IMHO, too 
unwieldy to recommend.  In this case, if I had a fast pipe or backup 
storage close to the client, the file systems would no longer be "large"!

I still like the IBM TSM model of backing up a file exactly once, 
eliminating all this full-backup nonsense.  (Of course, back on the 
backup server you shuffle files around as much as you can to maintain 
good restore times).

Hope this helps!   cheers, wayne

Martin, Jonathan (Contractor) wrote, in part,  on 6/12/2006 1:30 PM:
> It is my understanding that I can't "restart" a failed backup with 
> checkpoint enabled unless it is still in the backup window?  Is this 
> correct?  Also, does anyone have any tips on using checkpoint and 
> large backups?