Why Cloud hasn't made "going down" a thing of the past
Yesterday I read about how Foursquare had major downtime. They're a customer of Amazon AWS's cloud, yet they had downtime comparable to the conspicuous disaster in 2007 at 365 Main that took down non-cloud-hosted Craiglist, Yelp, Technorati and SixApart for up to day. What happened? Wasn't cloud computing supposed to solve this problem?
in this case, there was apparently a problem in the AWS datacenter that hosts Foursquare's services. However, as in the case in 2007, Amazon's downtime was far less than that of Foursquare. This brings up two myths of cloud computing: 1) Cloud computing never goes down; and 2) Your site will never go down if it's hosted on cloud computing. Both are terribly and dangerously false.
Hardware will always fail, suffer from misconfigurations, or otherwise be the victim of human failures, so from the user's point of view cloud computing will never reach 100% availability. In fact, when Amazon says that it's cloud is reliable, they speak of their entire cloud, not a particular end-customer's application hosting. The key is what your cloud computing provider does when the hardware fails. Amazon does very little, simply offering you the ability to restart your software on another machine. Others - such as us - will restart it for you. How quickly the restart happens, and how much demand on your software that it presents depends on the cloud you choose.
But that's only the beginning of true application reliability. As we saw with Foursquare yesterday and the 365 Main event in 2007, if your software isn't written and deployed in a way to be tolerant of failure, the time needed to bring it back up can be a major disaster for your business. This preparation for downtime is called DR, or disaster recovery, and people realized that the 365 Main event was highlighting that many companies haven't given it much thought. DR can start with as simple a preparation as writing your software to be resistant to database corruption caused by downtime or perhaps adding monitoring that shows whether it's working properly, or it can extend all the way to having a live "warm" standby site that can take over if your primary site fails. Cloud computing can make these options possible or affordable, but it does not guarantee them. So simply placing your site in a cloud doesn't guarantee uptime, even if it does put it in a datacenter that is professionally managed.
Every day, as i meet potential customers, read advertisements from other cloud companies, or catch up on cloud computing blogs, I've been making a mental list of the "myths of cloud computing" that I've been hearing from them. These myths are dangerous: they produce a mismatch of expectations between cloud customers and vendors that can injure everyone - especially the cloud user who expected that their cloud-hosted web site would produce far more professional results than it actually does.
A partial list of the myths I plan to explore are:
- Cloud computing never goes down (I'll follow up on this article in more depth and make it part of my Cloud 101 class.)
- Computing resources in the cloud are infinite
- Cloud computing is nearly free
- Cloud computing will make my software compliant with regulations and certifications
- if you have any more you'd like me to discuss, I'd love to hear from you.
Like any myths, there is a kernel of partial truth to these assertions, which is the reason cloud computing is so attractive. But how much of these benefits you actually get depends both on you and your cloud vendor.
TrackBack URI for this entry
