undefined | Better HN

0 pointsplatz7y ago0 comments

Ah, I didn't know Paxos was on the decline or that etcd uses raft

0 comments

3 comments · 1 top-level

espadrine7y ago· 2 in thread

I wouldn't call Paxos on the decline. It has several nice properties that Raft lacks[1]. However, implementing it correctly is more brutal than Raft, which is cut into easier-to-understand components.

Also, the paper introducing Paxos[2] is famously obfuscated, while the Raft website[3] trivializes the problem, and has ready-to-use libraries.

As a result, there are more recently-developed systems in Raft than Paxos, but I expect there is more volume hitting Paxos systems than Raft ones in the world.

[1]: http://rystsov.info/2017/02/15/simple-consensus.html

[2]: https://lamport.azurewebsites.net/pubs/lamport-paxos.pdf

[3]: https://raft.github.io/

evanweaver7y ago

As a member of a team that has implemented Raft in production (https://fauna.com) and in conversation with others who have, the general consensus is that Raft, Paxos, and even Viewstamped Replication are variants of the same fundamental algorithm.

There is an interesting paper about the commonalities in consensus here: https://arxiv.org/pdf/1309.5671.pdf

suj1th7y ago

Thank you for sharing an interesting paper. The similarity between Raft and Viewstamped Replication is easy to gauge. The original Raft paper too acknowledges this similarity. OTOH, the similarity between Paxos and Raft is not very apparent (to me). The paper in the link seems to predate Raft, and only considers VR, Zab and multi-Paxos.

j / k navigate · click thread line to collapse