Virtualizing your data protection - it ain't no cakewalk and takes a lot more than grabbing another copy of back-u-up 6.0

A colleague (Jeff Byrne) and I recently did a webcast about storage issues surrounding server virtualization. On the tail end of that session, we received a number of questions about data protection in virtual environments. That in and of itself isn't necessarily news - optimal virtual server backup is without a doubt confusing and somewhat frustrating. But the fact that there are always questions around protection, and that there are no clear answers applicable across every situation, is leading me to shift my perspective a bit.

The thing is, most practitioners have long relied on a tried and true approach to data protection that leads us to be a bit complacent. That tried and true approach is generally an out-of-the-box software solution for file-oriented protection of Windows environments. Sure, you might have an image oriented bare metal recovery solution too, and you're probably doing some snapshots here and there, and brick-level, etc., but the majority of the industry blankets the windows server infrastructure with file backup as well. Today, when it comes to block attached server virtualization infrastructures, this tried and true approach forces you into a lot of compromises compared to other potential solutions.

But mixing it up even further, is the enormous number of other potential approaches, including CDP-based or CDP-like approaches like Asempra, FalconStor, Datacore, etc., delta-based system imaging tools like Symantec BackupExec System Recovery Server, delta-based replication tools like InMage (actually replication and CDP) or even FalconStor again, file system or host snaps (ZFS, Vxfs, etc. - great post on ZFS snaps and VirtualBox rollback by Brian Leonard), array-based snaps, virtualized storage appliance snaps and replication ala LeftHand's VSA, host-based dedupe like Avamar, and more. Moreover, each of these technologies represents a compromise on some dimension and an optimization in another - including RPO/RTO, bandwidth consumed, IO and/or CPU load created, what you can do with data that is protected, and more. So the point is, backup for VM environments is not a single solution today, it is more like a continuum, and you might choose multiple products or technologies and apply them to different virtual servers, depending on what your goals, constraints, and protection service level requirements are for those servers. Backup is no longer a checklist item, and today it is more like an art. A check-list won't get you to an ideal solution, and no matter how ideal you think your solution is, it is likely a compromise in some dimension that you may not even be thinking about. In my opinion, anyone who is telling you different, is making a lot of compromises that you may not be aware of. And right here I'm talking only about backup solutions, not inclusive of backup devices - things get even more complex when you start thinking about where you put backup appliances (physical or virtual, ala FalconStor's VTL virtual appliance, even your CDP technology (InMage has some interesting tricks up their sleeve to put CDP at the port level in your SAN), or virtual devices within edge devices like Riverbed Steelheads or Cisco WAAS devices), and what type of appliances those might be (VTLs, D2D, etc.)

So as an aside, and backtracking just a bit here, the interesting thing is that your one tried and true backup approach (file oriented) may actually work, if you're willing to think about making your virtualization storage environment file based instead of block based. In light of the data protection challenges, NFS storage for virtual server images becomes pretty attractive. You get bang for your buck around data protection simplification, and there is lots of vendor energy focused on making NAS performance scale like never before - from the likes of OnStor, Celerra, BlueArc, Isilon, IBRIX, ExaNet, IBM and SoFS, HP and the 9100 xds, Active Circle, and many others. Frankly, it's about time for someone in this space to do a throw down against block based storage server virtualization and paint a picture of what the tradeoffs, on both sides, look like. So far, I don't believe anybody has stepped up, and it is about time to "put up" in a real world lab environment while taking a serious look at total solution TCO. More to come on this topic.

The good news around data protection outside of file-oriented storage, is that we (at Taneja Group) plan on putting some serious energy into helping you understand the spectrum of available solutions over the next few months, as well as stay on top of potential long term improvements so you don't waste today's dollar on band aid solutions when something better is just around the corner. Have ideas on specific challenges you think are missed by all of the solutions available to you? Drop me a comment.