September 24 2006

Amazon outages hit other retailers on the shared platform

eWeek reports in depth the server problems suffered by Amazon - and the matching issues at Target and Borders, companies hosted on the same platform. While Amazon's uptime is generally excellent, the recent glitches show the risk of having your operations tied to another's.

There's an inherent conflict between wanting to be based on a supplier's infrastructure for reasons of cost-effectiveness and requiring a separate, de-coupled infrastructure for resilience and independence.

Early Amazon adopters of their service model seem to have suffered recently by this close tie - as reported in detail at eWeek.com.

Amazon's servers' availability is normally at 98-99% (according to traffic monitoring firm Keynote) but dropped to the low 70%s over 20-21 September. Target and Border's performance suffered a similar, mirrored slowdown.

Now both of these companies were early adopters of the Amazon outsourced service and so one wouldn't wish to extrapolate this effect to all of Amazon's clients. Indeed, it's be astounding were the latest customers (eg our very own M&S) not architected to be fully de-coupled from server problems or component malfunction - in line with best practice for system architecture.

The issue with Amazon simply gives a high-profile name to a problem of shared services.

Most retailers are not yet at the level that would require Amazon Services as our partners (we may dream...) but increasinly we'll be using 'software as a service' for parts of our offering - analytics, email marketing, payment processing... A key aspect of good procurement will be to ensure that the business can function uninterrupted in the case of server overload or malfunction or where individual components are offline for a period of time.

Ian Jindal