The dangers of high availability

About 15 years ago I heard someone giving a lecture on a system he managed. He claimed they need to provide 5 9’s of availability, as it’s an extremely critical system. Needless to say, he didn’t deliver 5 9’s. Not even one. The system worked perhaps 79.999% of the time. Not very surprising. But the real surprise here is that this level of availability was good enough.

People tend to think their applications require extreme high availability for a wide variety of reasons. However – most don’t need high availability.

Now, since high availability is so costly to implement (hardware costs, architectural and development costs, etc), I think you really need to ask yourself – does my application really need high availability – or am I taking all this effort just to soothe my ego?

There are levels of high availability, and think well before you answer. Do all of the users need access to the application 24*7? What are their working hours? Can some users experience failure while most users still work? Can users experience partial data loss (restart a wizard, or re-filling a form)?

I, for one, think that giving a good enough availability, and have a very good monitoring solution with someone to solve crashes on a 30 minute notice is probably good enough – and is much more cost effective.