Recently, a whole slew of tweets showed up across my feed dealing with the perceived and measured issues across multiple public cloud providers infrastructure.
One of the posts comes from Chris Hoff (this post in particular) that describes quite clearly what the differences are.
Service providers (anyone doing cloud services, virtualization, colocation, bandwidth, whatever) live upon the idea of over subscription. We make our revenue banking on the fact that not everyone needs their full allotment all the time.
With over subscription there is a chance of reaching a state of over capacity. Anyone using this business model needs to understand that they must be ready for it to happen. It isn’t an issue of ‘will’ but an issue of ‘when’. Good engineering can keep the ‘when’ at bay, virtually forever, and that is what you need to be prepared for.
Another recent posting via The Register (featuring @GeorgeReese) had some data dealing with network latency within the Amazon EC2 network. I don’t have any opinions about what was in this article, but it is something that is going to come up again and again as this new model of computing (for the masses) solidifies and grows up on the Internet.