[agazso]

For anyone more deeply interested in this topic I recommend to read this blog post from Archagon. It describes the different alternatives (OT, CmRDT, CvRDT, diff sync) for writing a collaborative editor. And unlike academic papers it is written in a format how a programmer does research and thinks about a problem in real life, so it's very natural to follow, even if it's long and complex.

http://archagon.net/blog/2018/03/24/data-laced-with-history/

(I am not affiliated in any way, just enjoyed it very much)

[dboreham]

Alexei's work on this subject is very good: the best practical introduction to the subject.

However, I feel that it is worthwhile (necessary?) to completely understand the academic basis for CRDT. A good place to start is Shapiro et al's second paper : https://hal.inria.fr/inria-00555588/document .

In order (sic) to understand the paper you need to have a grasp of Order Theory. This is not terribly hard to get your head around. This is a good place to start: http://jtfmumm.com/blog/2015/11/17/crdt-primer-1-defanging-o... also https://www.wikiwand.com/en/Order_theory and basically stop when you understand this : https://www.wikiwand.com/en/Lattice_(order)

The reason why the formal basis is important is: that's the whole point of CRDT -- previously (been there, got the t-shirt..) folks just made up replication mechanisms they thought would work. Then they build them and embarked on a process of fixing the bugs. Sometimes that took decades. CRDT is nothing new in terms of : software to perform eventually consistent replication. The new thing is that there's a way to formally prove that your bright idea for replication will in fact work (as in : it will converge and it will have the consistency properties you expect). So if you're not seeing that aspect, then you're really not with the program.

btw originally the C stood for Commutative or Convergent, not Conflict-Free. There are plenty of CRDTs that cope with conflicts consistently, rather than being conflict free (e.g. LWW Register).

[aboodman]

FWIW, I disagree that it is necessary (or even desirable) to understand the academic basis first.

The core idea of CRDTs are intuitive and easy to understand from an engineering perspective. For me, the academic literature unnecessarily complicates and obscures what's going on, and I would also point newcomers to something like http://archagon.net/blog/2018/03/24/data-laced-with-history/.

It probably has to do with where you are coming from. If you're not a mathematician or theoretical computer scientist, I don't think that reading these documents is going to be a very helpful start.

[dboreham]

I hear you because I thought the same way until recently. I have worked with eventually consistent replication since long before the CRDT papers and thought initially it was just some academic mumbo jumbo layered over intuition. However, based on my journey through the field I'd encourage you to try again to grokk the academic side of the work because it's really the key insight and (for me) it's very much not intuitive. In fact that's the tripping hazard in all this: the simple examples are intuitive which leads to a sense that everything is similarly intuitive. Not so, for me at least.

I'm neither Mathematician nor Computer Scientist (I'm an Electrical Engineer), but I have learned not to fear formal notation and advanced abstract concepts. They're really not that hard to understand once the terminology is decoded.

[sophacles]

At one point I was strugging with the basics of CRDTs. Someone linked me to this gem: https://www.youtube.com/watch?v=pMMDVphop40

It made a lot of stuff click for me. HTH someone!

[fizwhiz]

what t-shirt?

[dboreham]

https://en.m.wiktionary.org/wiki/been_there,_done_that,_got_...

[appwiz]

Most recent submission to HN about this paper: https://news.ycombinator.com/item?id=21405583