Alan Greenspan once characterized the dot-com bubble of the 1990s as "irrational exuberance." To me, much of the rush to the cloud is the same. In our excitement we forget all the lessons we learned, often the hard way, in our own data centers. A wonderful example of this is cloud-based storage provider Nirvanix. Founded in 2007, it announced in September that it was discontinuing its services. Users had two weeks -- yes, just two weeks -- to retrieve their data.
Enterprises are notoriously slow to change, so finding a new cloud storage provider in 14 days would be incredibly taxing by itself. On top of that, bandwidth is an issue. Assuming that you could get 1 Gbps of bandwidth out of Nirvanix and into another provider, moving 100 TB of data would take 9.5 days. Getting a reasonable amount of bandwidth is a big assumption, too. Like all cloud providers, Nirvanix probably relies on oversubscription of its infrastructure, making this problem the modern equivalent of a run on a bank. With everybody trying to retrieve their data simultaneously, nobody succeeds, and Nirvanix does not have the capital, the will or likely even the staff to do anything about it.
Back to IT best practices basics
Why is this such a problem? For starters, most cloud services lack a decent exit strategy, often quite intentionally. Take Amazon Glacier, for instance. It's free to transfer data in, but can be terrifically expensive to transfer it out. Most enterprise IT shops call that vendor lock-in and manage it. But with non-IT folks calling the shots when it comes to cloud services, many of these lessons never get applied to the new technology, or even considered until it's too late.
Another problem is redundancy. Enterprise IT builds redundancy in from the application level all the way down to the power feeds. With storage, for example, data is usually replicated to another array, often off-site. If data is important enough to keep, it's important enough to copy.
So why do organizations choose only one cloud storage provider? The same enterprises that are fanatic about redundancy in the data center turn to the cloud and place all their proverbial eggs in a single cloud basket. With high-profile outages occurring almost yearly to Amazon Web Services and Microsoft Windows Azure, there is growing evidence that cloud providers are not infallible. Even if their technology is solid, there is no guarantee the business side is, as in Nirvanix's case. Even Google routinely displaces users by discontinuing products.
If we apply the IT best practices from in-house storage to the cloud, we'd have redundant copies in separate providers, and the shutdown of one provider would merely be a big inconvenience, not a life-or-death scramble.
Pundits decry this whole Nirvanix incident as an example of where the cloud fails its users. But I see it differently. The cloud may have failed the customers, but once again IT failed their organizations. In their irrational exuberance, in the rush to turn over a new IT leaf via the cloud, organizational management forgot, or ignored, all the good things we've learned from our own local data centers. It is age versus wisdom, and the cloud teenagers need to take off the headphones and listen to the IT elders. Vendor lock-in and exit strategies are real issues, whether with private or public cloud. So is availability, protecting not only against the traditional technological failure but also the new concept of business failure. What happens when a company goes bankrupt? What happens when a company discontinues a product?
The cloud is a revolutionary tool, but it isn't exempt from failures. And its users are not exempt from using their heads.
About the author:
Bob Plankers is a virtualization and cloud architect at a major Midwestern university. He is also the author of The Lone Sysadmin blog.
- Disaster recovery in the age of cloud –ComputerWeekly.com
- Essential Guide to Disaster Recovery –SearchStorage.com
- Cloud Backup and Cloud Disaster Recovery Essential Guide –SearchStorage.com
- A Guide To Leveraging Cloud Services For Backup And Disaster Recovery –Iron Mountain