A few months back, when VMware let the cat out of the bag that it would be building its own public cloud, it said that it had 480,000 customers with an estimated 36 million virtual machines running in their data centers. On Tuesday it officially launched the the Hybrid Cloud Service at its Palo Alto headquarters and explained …
sell it based on resource pools, let the customer scale CPU or memory however they want. That's the whole point.
Also no mention of what storage they are using.
Amazon pricing is absurdly high (as is most other cloud providers, haven't seen one that is competitive yet). A large reason for this is average utilization in such an environment is pitifully low - on part with what people had before the era of virtualization. The problem is compounded by fixed instance sizes.
It's pathetic if vmware can't leverage it's own technology of pooling resources and being able to oversubscribe in it's own public cloud that it's customers have been able to do for many years in their private installations. I say this as a vmware customer for the past 14 years ( pre 1.0).
This cloud seems similar to what the likes of Terremark(now Verizon) have been offering for probably 4-5+ years now (using VMware). Though Terremark at least offers resource pooling in their enterprise cloud offering (though the pricing is crazy high as well).
For the apps that I've supported over the years(several of which have been built from the ground up in the past few years, I'm not dealing with legacy crap from the 90s here) at least and I think this experience is somewhat typical - "extending" the application to burst to an external provider of any kind is far from trivial from the application layer (the infrastructure which this provides is the easy part).
Single points of failure for databases, caches and other things are quite common(and fixing such things is often very complicated), and all of a sudden shifting some load to an off site provider and having to eat the latency between that and other local systems not to mention the configurations involved from an application layer...just doesn't make the concept very feasible for *most* things. There will be exceptions for certain types of apps that are built from the ground up to do such sorts of scaling but I believe they will be very few and and even farther between.
Also par for the course is the developers/product managers having no interest in fixing such problems, they'd rather get the next whiz bang feature out the door that the users can tweeter about.