« Instant VMware Backup - Edited | Main | Editorial - Backup Licensing »

June 12, 2008

Comments

Feed You can follow this conversation by subscribing to the comment feed for this post.

Scott Harney

One place where operational backup to tape is not quite dead yet is high rate of change databases (Oracle). LTO4 and similar tech (STKs T10000) can do 120MB/s . And that's not just a benchmark, though sustained real-world throughput in my environment is more typically 85-95Mbps.

De-dupe, source or target, doesn't buy you much with these types of backups. Though if you can describe a scenario where it does, I'm all ears. But keep in mind tape media is still < $.20GB Though that figure is media alone. so there's a grain of salt in that.

If you're able to feed tape at high enough speeds, at that cost, even disk-to-disk-tape doesn't make a lot of sense for this type of backup. target based de-dupe may help depending on the app, but that would be really hard to figure out for a backup engineering type.

The big caveat in the straight to tape scenario is how fast can you stream to that tape? If the database and frame are busy working with a high rate of change, you may not get that kind of throughput so you end up multiplexing which is fine, up to a point.

And I'm a de-dupe believer. For VMWare, file servers, app servers, it is pure gold. But there's still a place for tape as near as I can see it from the customer side. And not just for long-term compliance archiving and DR off-site vaulting.

Scott Waterhouse

All those points are true. I do think however that the total cost of ownership has to take into account tape libraries, tape drives, floor space, etc. And at the end of the day I think we have yet to see any sort of concrete study that takes an impartial look at the TCO of disk with deduplication vs. the TCO of tape. If you want, I think you can add some soft costs to the discussion, like reliability, performance, possibly reduced administrative effort, greater granularity for recovery, and so on, but I personally prefer to stay away from those. They should be the "icing on the cake" in terms of making a business case. Also, my experience is showing that larger sites are willing to pay a modest premium for backup to deduplicated disk rather than tape. Not large, but modest--10% maybe?

As far as high change rate dbs, I am with you... sort of... "High" is subjective. So what I would say would be dbs with a change rate in excess of 10% per day and a relatively low compressability (less than 2:1, maybe less than 3:1). Basic hardware compressability (same sort as you get with tape). If your db has more change than that, or less compressability, you may be better off going to tape. Actually, you may be better off going to straight disk or VTL (no deduplication) if your retention period is sufficiently short. But given a "normal" retention of at least 30 days plus monthlies, I suspect that there is still a good, or good enough, cost case to be had. The quick math yielded a 5.6:1 deduplication ratio under this scenario, for what its worth.

Finally, I think one of the things that will help the business case a lot here (especially for the dbs we are talking about) is replication. If I can replicate my deduplicated data over IP and only use 5-10% of the bandwidth I would need to replicate a full backup, that is pretty compelling. Particularly if that means no more couriers, storage fees, recall fees, no hassle with implementing tape drive encryption, etc.

Scott Harney

Right. The $.20G doesn't include all the true costs in a large scale tape library. For biz case purposes, the other soft benefits are generally stated separately.

By High, I do mean potentially higher than 10%. Of course that can vary wildly in a 30 day period depending on the database. But if you can't stream from your source faster than say 40MB/s sustained, you're not getting much benefit from your high speed tape drive targets or VTL targets. So if you've got a substantial existing tape investment, it's hard to make the biz case to replace it wholesale if a significant amount of your environment consists of scenarios like this. Of course costs are going down every day. Working at a relatively conservative site, in a conservative industry, getting fully off of tape and onto VTL for most operational backup data will likely take a while.

The DR scenario with replication is definitely compelling. The costs of vault vendor and tape handling vs. that circuit cost can be made. particularly if you're doing lower bandwidth b/c of de-dupe. If you can refine your DR strategy to something realistic instead of just dupe everything with no prioritization whatsoever you can probably still make it work.

We are a networker environment, primarily with tape targets. We are implementing Avamar within the next 90 days to help with a recent VMWare ESX deployment and some other backup clients that are tailor-made for an Avamar scenario. This was an easy business case to make. If I were on the other side (sales) this is a good way to get the customer moving in the direction of getting off tape for operational backup. These are easy targets for non-tape backup and can prove the case. There's still a LOT of variability and change on the target-side de-dupe and VTL front. We're watching with interest. It's nice to see some excitement in the normally dark and murky world of backup and recovery.

I don't find data at rest tape encryption too difficult. And I've done DR with it as well. And it does not affect throughput at all. For compliance purposes, VTL vendors may want to consider doing some sort of hardware-assisted backend encryption to their VTL volumes.

The comments to this entry are closed.

Search The Backup Blog

  • Search

    WWW
    thebackupblog