[yshalabi]

Yeah, my question was more of a curiosity one regarding the utility of EC for applications. Sorry if it seemed like a brazen dismissal of your article. I am actually waiting to read your wait-free removal and lock-free append extensions.

Now I understand your first paragraph a bit better :P. I would avoid using EC. Instead I would find another term. To me, EC implies that the object can be inconsistent for some period of time but will eventually reach a consistent state across all nodes. Your titling of the link and your use of the EC term kind of threw me off a bit.

"Modern computer are eventually consistent systems by virtue of the caching layers (and local registers) - only once modifications make it to a shared cache are they globally visible."

I agree with you but I would leave registers/cache out of this. Yes, if a read to a memory location could return stale data then this would be absolutely be a problem. But I would argue that reading a memory location into a register then performing RMW ops on it does not necessarily put the system in an inconsistent state. This because RMW ops on a register dont change the actual shared object.

I think a better example for what you are talking about is write buffers in individual processor cores. These can contain intermediate state visible to that processor but not to others. Such relaxed consistency procesors will require appropriately placed fences to achieve linearizability.