100% availability - No Problem

Don't Oversell Availability, warns Peter Warren.

You've maybe never heard of S3, but the chances are you've used a website that uses it either wholly or in part in the last month.

S3 is Simple Storage Service - web content hosting on steriods, and is provided by Amazon. It is used by services like the excellent Twitter, Smugmug and Pownce, and thousands more. It's key selling points are its low cost and high availability - and at the end of last week it suffered a fairly substantial amount of downtime.

As a user myself, I've been surprised by the reaction of the S3 user community who view this like it is the end of the on-line world as we know it. In fact, the service was down for less than 3 hours.

Many users seem to have taken the marketing speak about "ensuring that the data will always be available when you need it" and the reassurances about redundant copies of data on different servers (the much hyped 'cloud') at face value - and have taken 100% uptime as the minimum that they should expect.

Personally I always view these claims in the same way as claims like 'this ship is unsinkable' - it always seems companies lack the imagination to understand where the critical weak-point is. In this case, S3 may (according to some rumours) have suffered a good old-fashioned Denial of Service (DoS) Attack - its authentication server got a gazillion requests and could not keep up. The 3 hour downtime was how long it took Amazon to create extra capacity.

My complaint is really about the way Amazon dealt with this, and I think this might be what is at the root of some customer ill-feeling (some customers seem to be downright unreasonable though).

I got a call in the small hours of the morning to say our website was down. Working from home it was clear that some important files hosted on S3 were inaccessible, but there was nothing from their Helpdesk that I could find on the support site to say either that Amazon knew something was up, or if they did know when it would be back.

This meant I had little of value to say to our own customers.

It took a note on the user forum before we got any word from Amazon. Frankly that's not good enough.

Many users are now asking for service status information. I hope they provide this soon.

I guess the moral of the story is: don't oversell availability, practice your customer response for when the inevitable unavailability occurs, and be very open. After all, I have Service Status pages, why don't they?

Peter Warren is a guest blogger. 

Tags: 

Categories: