> The second question is equally important:given a monotonic specification for a problem, how can I imple-ment it in practice?
Can anyone explain a good way to use these results?
The OP (acolyer.org) seems to suggest that this boils down to using CRDTs but that doesn't seem to be isn't really the answer but I can't figure out how to use/properly think about the results in this paper (as opposed to others that are a bit more practical). I've crawled through a bunch of paxos papers[0], and at this point it seems to that the distribution part of the problem is somewhat solved but the coordination required is what we're trying to attack now.
Whenever I see systems that use CRDTs it's usually either a K/V store or a concurrent editing application. Also, most articles will gloss over the fact that both types of CRDTs (cmRDTs and cvRDTS IIRC) are actually roughly equivalent (pure operation-based CRDTs[1]). IMO pure operation-based CRDTs are just distributed WAL by another name -- and if that's the case then we already have the solution for this (paxos-like quorum on the WAL segements).
All this boils down to the thought I've been holding lately that distributed consistency isn't hard in this day and age -- WAL (which you generally want for hard-disk-level consistency to start with, if you're not just waiting for big fsyncs) + synchronous commit (whether quorum or all nodes) is all you need to achieve a perfectly consistent distributed system, but what is hard is doing that with low latency and high throughput.
[0]: https://vadosware.io/post/paxosmon-gotta-concensus-them-all
[1]: https://www.researchgate.net/publication/320371248_Pure_Oper...