Because nothing is certain 
A couple of weeks ago, Ray Lucchesi over at Silverton Consulting wrote a blog about whether or not cloud storage needs to be backed up. Ray makes a number of good points; it’s worth a read. I’ve been digging into the topic a bit on my own and, like Ray, I think there needs to be more discussion on the topic.
Most cloud storage service providers have well defined service levels that can easily be found on their web sites. The one thing they all seem to have in common is that they discuss service availability, not data availability. Take a look at some of the cloud storage vendor SLAs – I haven’t found any that even discuss backup or data availability. You buy capacity, not functionality. The most you get is remote copies for DR in the event of a geographic failure – but restore of deleted files or recovery from data corruption has to be handled by the application, not the cloud storage provider. There are no point-in-time copy or versioning options.
Losing data is bound to happen at some point and creating multiple copies of data at different locations doesn’t protect users from data loss that happens from acidental deletes or data corruption; it just deletes the data in multiple places and mirrors the corruption! In today’s cloud storage environment, there is not yet a concept of data protection – all that the consumer purchases is raw capacity – protecting it is typically up to the consumer. This is fine for some tiers of data, but needs to be a core consideration when deciding what can be stored in a public cloud versus building a private cloud and incorporating more functionality (at the corresponding higher price).
I understand that use of big-iron, sophisticated enterprise storage arrays was one factor that helped sink SSPs earlier this decade and that the more functionality that’s added in cloud storage, the more costs add up – but costs can be made up in other ways. Steve Duplessie recently wrote about cloud economics and scarcity – in his blog posts, he makes the point that cloud is a perfect model because you only pay for what you use: you have 100% utilization rates! Take power, cooling, floor space, and storage management out of the equation, remove capital outlay for future as well as current capacity, and I’ll bet that, even with enhanced functionality, cloud comes out cheaper.
Cloud strorage provider SLAs translated: you can get to the storage 99.9% of the time, but there is no guarantee the data you put on that storage still exists! While I agree with Steve in that the cloud model provides compelling economics and will eventually win out, we’re still really early and offerings need to mature before the market gets there.
Related posts:
- Cloud Storage Enablement: Key to Commercial Cloud Storage Adoption
- Cloud Storage to the Rescue
- Breaking News: Microsoft Lifts the NDA on Azure Cloud Storage Services SLA!
- Real-time compression of primary data? You’ve got to be kidding.
- Nasuni’s Quick and Easy Hurricane DR Offer
Tags: Cloud Storage, Data Availability, Service Availability, SLAs
blogs
Even worse, you can send data to a cloud little by little in some type metered fashion (isn’t this great and its only $0.15/GB!?) but by golly if you totally lose your on-site data, how do you get 1TB, 5TB, 10TB+ of data sent back (and sent back ASAP)?
———-Thanks Rich – good point, we’re actually seeing a the return of sneakernet, shipping loaded drives instead of tapes. Seeing it both ways, to return data in the case of a local event and for cloud ingest! Ingest is another point we don’t hear enough about. Shifting a bunch of long tail static data from internal to cloud can be a big effort, so we’ll need something to help with ingest.
I agree wholeheartedly. What some Cloud providers have told me is that Cloud storage is more for static or reference data not Tier 1 primary storage.
But then what is cloud computing going to use for storage if not cloud storage. In this case it is imperative to have some sort of data protection strategy in place for your cloud storage. The sooner everybody realizes this the better.
Ray
—Thanks for the comment Ray – based on protection levels it is for maybe a secondary copy of static data. If the primary is out there, without some type of WORM capability, it is still exposed to data corruption or accidental delete, and the corruption or deletion hapens for all copies in a mirrored environment no matter how many mirrors you have! Lots of cloud storage has been developed as consumer-grade, enterprise-grade will require some work.
Very astute points and comments. All the more reason for cloud-hybrid data protection solution that includes deduped disk repository on-site for must have local copies (backups, CDP, replication, DR, etc.) and managed/tracked version copies going to the cloud, which is ideal to trickle copies real time as needed, and not interfere with backup windows.
Fast restore for most likely scenarios comes from the on-site repository. But a complete “friggin-thermo-nuclear-take-me-out-to-the-ballgame” type restore would require either a big pipe to the cloud (the kind Morpheus uses in The Matrix) or a long lunch (about 3,000 martini’s). And like any repository anywhere, if that long term stuff in the back-end cloud ain’t there, you may need a few more martini’s . . . .
TC
http://www.cofio.com
——Pour one for me Tony – when all is said and done hybrid looks more and more like the vable solution for a lot of the issues we’re facing! Thanks for commenting!
[...] blogged in the past about cloud storage data availability, or more accurately lack of data availability SLAs. Dropbox [...]
[...] blogged in the past about cloud storage data availability, or more accurately lack of data availability SLAs. Dropbox [...]