By relling on Oct 24, 2008
This week, several colleagues and I were discussing problems posing people who run datacenters due to the increasingly high churn rates in hardware. Some customers prefer systems which don't change every other month, so they can plan for longevity. Other customers want the latest and greatest bling. Vendors are stuck in the middle and have their own issues with stocking, manufacturing, plant obsolescence, etc. So, how can you design your datacenter architecture to cope with these conflicting trends? Well, back in 1999 I wrote a paper for the SuperG conference which discusses this trend, to some degree. It is called A Model for Datacenter.com. It is now nearly 10 years old, but seems to be holding its own, even though much has changed in the past decade. I'm posting it here, because I never throw things away and the SuperG proceedings are not generally available.