The Backup Blog

NetApp NearStore VTL: Now Guaranteed to Cause Data Loss!?

Edit: Note there is revised math in the comments section.

Well, the bloggers at NetApp have been threatening to introduce deduplication for their VTL for a while now, so it should come as no surprise that they finally did it. And just like that, they have assumed the unenviable position of being the last major vendor to announce a dedup product for backup.

(Just a quick role call here: EMC. Check. HP. Yep. IBM. You bet. Sun. Check. Dell. Roger that. Data Domain. Naturally. Sepaton. Yes. Falconstor. Yes. So NetApp, welcome, at last, to the game.)

Naturally, I had a lot of questions when I heard the announcement. Would it do replication? Is it a variable or fixed block scheme? In-line or post-process? How does it protect the hash index? And, most importantly of all, did NetApp decide to stick with VTL RAID rather than moving to RAID DP?

The last question is really the most crucial.

Deduplication and Replication

Some time ago I wrote a post Peace, Love and Revolution about the issues that replicating deduplication appliances causes with backup applications. Some six months later, I thought it would be good to update the state of the nation and see what progress has been made.

To quickly refresh, the issue is this: how do you restore data from a deduplication appliance that is the target for replication--when it is the appliances that is managing the replication process and not the backup application. (And to be fair, this is not an issue unique to deduplication: the exact same problems apply if you are just using virtual tape. The issue just happens to be more pressing with deduplication because if it is "just" virtual tape, I can use the backup application to manage replication and not cost myself any bandwidth; with deduplication, if I use the backup application I miss out on the 95% bandwidth savings that deduplication replication promises.)

DeDup Ratios and Fixing Backup

Normally, very high deduplication ratios are something to be pretty excited about. Particularly if you are the one getting them. If you get 450:1 deduplication, you are probably a pretty happy camper. It probably means you have saved yourself a pile of time, money, and administrative effort. It probably means your TCO is lower, your risk is reduced, and your backups are a lot more reliable.

So, life is good, right?

Well, maybe. But maybe not. Sometimes extravagantly high deduplication ratios are a sign that something is wrong. Really wrong.

I want to take the opportunity to talk about why, if you are getting 450:1 deduplication ratios, you may want to make some changes.

NetApp NearStore VTL: 4000 Times as Dangerous as an EMC EDL

I have previously discussed one of the big problems with the NetApp NearStore VTL here and here: it uses a type of RAID that is more dangerous than RAID-6 (including RAID-DP), and is, in fact, more dangerous than RAID-5.

Now I was content to describe the NearStore VTL as merely "more dangerous" than EMC's EDL at the time. After discussing the issues with Alex McDonald on The Missing Shade of Blue, I think I came to the conclusion that:

1. The NearStore VTL is, basically RAID-5. Yes, it is called VTL RAID by NetApp marketing, and yes there are some differences, but fundamentally, the risk and rebuild times are equivalent to RAID-5 (unless you aren't actually using the disk volume!).
2. In some respects, VTL RAID is worse than RAID-5 in terms of risk of data loss, as it tends to disperse some virtual cartridges across more than one RAID group. Meaning that the failure of a RAID group can impact a greater number of virtual tape cartridges than its capacity might indicate.

All this lead me to characterize VTL RAID as having 135% to 270% more risk of catastrophic data loss due to double disk failure than a RAID-5 volume of the same size. And all I was further able to conclude about the risk of a NearStore VTL vs. an actual EDL was that that the risk of a NearStore is much greater than an actual EDL that uses RAID-6.

It appears now however that we can be very specific about how much "much greater" is when it comes to the risk of a NearStore vs. an EDL.

According to the latest data cited by Alex: "Over a period of 5 years, RAID-5 is 4,000 times more likely to suffer from data loss than RAID-DP." His emphasis. Quote source is here.

HiFn Dedup Card

Interesting little bit of news being reported over at Byte and Switch today: HiFn has released their data deduplication co-processor card. What the story is missing however is the first thing that will be on everybody's mind: how fast is the silly thing?! And no, I don't consider "unprecedented 'de-dupe' performance" a meaningful characterization!

A little digging gets me here on the HiFn site: the spec sheet for the HiFn DR255 card. OK, so a bit of reading gets us to the key stat: 250 MB/s. Well. Color me underwhelmed. An EMC DL3D 3000 can get to 400-600 MB/s or better, depending on the circumstances. (In the spirit of full disclosure, some of this performance is provided by a HiFn compression card.)

Now, in HiFn's defense you can put up to 4 of these things in a single system. So the capability is there to get better performance on aggregate. However, that opens up yet more questions: how is the clustering and/or coordination handled between multiple cards? What is the cost of these things, and will it makes sense to put 4 in a single system, or does it just make more cost sense to have 2 systems that can provide equivalent performance? (Particularly as the need for a solid state disk is alluded to at one point.)

Lots of interesting questions, and not a lot of answers at this point. But it certainly points to the ongoing innovation in the deduplication space, and how there is lots of room at this point for different approaches to the problem. I will be very interested to see where (if) the HiFn approach works and where it will find itself in the marketplace.

Mozy News

Just a little piece of news to kick of the "back to business" postings after my absence. The folks over at Mozy have released Mozy Pro for the Mac recently. Mozy Pro is the business oriented version of Mozy, and offers some additional functionality that appeals to business users. Specifically, versioning of files, an admin console, application support specific to Exchange and SQL, and network drive support are extras of the Pro version of Mozy.

Being a convert to the Mac platform myself, I am happy to see the progress of Mozy in the Mac space. Now we just need Mozy for the iPhone.

All this on the heels of the news that Mozy has more than doubled its user base in the first half of 2008, as well as picking up support from Staples and Lenovo.