Can Backups Be Eliminated?

There seems to be much discussion around backups lately. One of the big issues is if legacy backup platforms stand a chance against new dedicated server virtualization solutions, but what I'd like to discuss in this blog is if the entire backup process itself can be eliminated by the intelligent use of snapshots, replication, compression and data deduplication.

Elimination of backups really comes down to two key issues: can you technically accomplish the goal and do you really want to accomplish the goal? On the technical side, I'd have to say that we are pretty much there. Primary storage is becoming more resilient and as primary storage vendors add the data services I mention above, I think technically the job can be accomplished. Many systems support an unlimited number of snapshots and/or copies of data by leveraging deduplication. Most have data replicated to a remote site, so you are covered from a single site disaster taking out all of your backup along with your primary store.

The first challenge to this approach is that you are counting on a single meta-data table to keep track of all the data interrelations. If that table gets corrupted, then most if not all of your point in time copies may not be able to be read. The other challenge is that if the whole storage system fails, if your redundant copies of data are all on the same system, you have a problem.

I know that the chances of either of the above scenarios happening is relatively small, but there is a chance, and isn't that what backups are for? Cover yourself just in case something unlikely happens? Also, if you did go with a replication option, it should help you recover from those two situations, but you do have to have a way to get that data back. If it is an entire restore across a WAN connection you may be out of business before the recovery completes. While you could replicate to a second unit locally and then a third in DR, doesn't that begin to sound just like a backup?

A situation that could occur and one that replication would not protect you from is if the deduplication or snapshot engines produce a silent error that does not appear right away and somehow the deduplication or snapshot engine reports that it is working correctly. If this happens, you may not know you have a problem until months later.  I have yet to find this occurring in an environment and have only heard anecdotal stories of even the possibility. Also most deduplication processes have self-check code to help prevent it from occurring, but it's better to be safe than sorry.

There are steps you can take to reduce the exposure to any of these failures even if the likelihood of them actually happening is slim. Using these methods as a primary and even secondary recovery point is certainly acceptable. For me though, there is something comforting in knowing that your data is on a separate platform (disk or tape) managed and protected by a separate process.