Last time out, we went over the reasons why Avamar is different and faster than other backup applications for backing up unstructured data sets.
This time out, I thought we should turn our attention to NAS (filer) backup under Avamar. Again, we see big differences in how Avamar handles NDMP backups. Again, we see that Avamar backups can take only 10% of the time of a standard NDMP backup. But it turns out that the reasons for this are totally different than the reasons that Avamar is so fast at unstructured data backups.
First, a bit of background. Avamar doesn't really do full or incremental backups. Avamar does a backup. That backup data is written together with existing backup data on an Avamar server in a way that a full system image is present for every day on which a backup occurred.
We don't call this a synthetic full (because that implies a bunch of server side activity that actually never takes place with Avamar) but the result is pretty much the same: you have a full image of the system for every day you have a backup. If you back the system up every day of the week, then after a month you will have 30 "full" backups on the Avamar server. (Actually, this tends to be better than some implementations of synthetic fulls which destroy the previous full image when creating the current full—Avamar's implementation is non-destructive in that sense.)
As a nifty bit of trivia, if you were to look at the file system on an Avamar server, you would actually see something like /backups/client x/date; where each client gets it own sub-directory, and each date on which a backup occurs is in a sub-directory. Mount the directory, and do an "ls" and you would get the complete contents of that system for the date of the backup (which, with the appropriate permissions, you could then copy directories or files out of).
Back to NDMP. When we do an NDMP backup, the first backup is an NDMP level 0. The NDMP equivalent of a full backup. But subsequently, Avamar is able to employ the backup technology described above to do level 1 (NDMP incremental) backups and only level 1 backups.
This turns out to be a really big deal. For two reasons.
First, it means no more Level 0 backups. Level 0 backups are painful, because they can take a long, long time. On a bigger filer, it wouldn't be abnormal for a Level 0 backup to take 3-5 days. Do a Level 1, and it is a matter of a few hours, or half a day. Big difference. Tell your business that their production systems are slow for 5 days because of backup (3 of those days being weekdays): not good. But if the production filer runs a little slower over the weekend during a Level 1: not so bad. Eliminating Level 0 backups is a very good thing because it can save a huge amount of time.
Second, it means that restores aren't compromised. The basic reason why nobody else can do this is simple. If you do a Level 0 backup, and then successive Level 1s, if you have to restore the whole filer, then you have to restore the Level 0, and then restore the individual Level 1s successively after that. This isn't so bad if you have done a Level 0 on the weekend, and only have, at most, a week's worth of Level 1s to restore. But if you tried doing a Level 0 once a month, you could have 30 Level 1s to restore on top of this to get your filer back to its last known good state. And if you are using tape, you could raise a child, train a dog, or take a long break and do some good deeds in the Peace Corps before this process is finished.
But not with Avamar. With Avamar, I just restore my last image. It doesn't matter that I only backed up using an NDMP level 1, I still get a "full" image to restore. With Avamar, every restore can be done as if restoring from a Level 0 backup.
And, how long do you think we can keep this up? As Florentino said: "Forever."
Incidentally, further discussion of this can be seen in this white paper by The Evaluator Group.