Hacker News new | comments | show | ask | jobs | submit login

I'm afraid you're not quite correct. CAP says that, in the face of potentially arbitrary network partitions (which are precisely modelled by a set of dropped messages) you can be 100% consistent, or 100% available, but you can't be both.

If the network is allowed to drop packets there are times where either you must not respond to requests (as doing so would violate sequential consistency) or you must respond incorrectly, potentially with stale information.

The network partition that forces you to drop one of these guarantees might be quite dramatic - but note that, for example, a quorum system is only available on the majority side of its partition. Therefore if a single node is unable to deliver messages (due to a network partition event), it will not be able to correctly respond to requests and must either not respond to its clients or respond incorrectly.




> I'm afraid you're not quite correct.

So then what am I missing? It sounds to me like the rest of your post is reiterating what you just called incorrect.


You said that wasn't what TFA was describing, but it is.




Guidelines | FAQ | Support | API | Security | Lists | Bookmarklet | DMCA | Apply to YC | Contact

Search: