Amazon's EC2 "cloud" crashed really hard this week. Reddit has been either read-only or paying-members-only for almost two days now, and dozens of major web sites including NYT have also been affected in one way or another.

Linode, meanwhile, keeps rockin'. That there, is the proper way to do "cloud".

To be fair, it's not entirely Amazon's problem that those sites are having issues. Those sites didn't have any geographic redundancy; they put all their eggs in one basket. Netflix, on the other hand, suffered no ill-effects, since they planned for this.

Amazon openly advertised that if you purchase EC2 & EBS instances in multiple "availability zones" ( == datacenters) even in the same "geographical area" ( ~= state), you would be safe from all but the most catastrophic natural disasters.

Many sites probably relied on that promise. Reddit for example had servers in multiple availability zones in VA. Turns out Amazon couldn't deliver what they advertised, as multiple availability zones went down at the same time last week.

It is impossible to completely isolate two (or more) systems which are as closely interconnected as EBS availability zones. This can probably be traced back to some fundamental law of physics. If one system of such a system-of-systems disappears, there must be some sort of impact on the remaining system(s).

Put succinctly, it's an active system, so there's going to be one or more poles in the right half of the s-plane. -rt (galaxies have time constants too)

_________________

/* TODO: need to add signature to posts */

My understanding (and what has been reported in the media and by Amazon) is that there can be multiple availability zones in the same datacenter, so proper geographic diversity can't be achieved through multiple availability zones alone.