To me it's clearly simpler. Paxos involces special client numbers (to guarantee ordering), but you still have to fall back to randomized delays to avoid livelock; raft just starts with randomized delays up front.
If you can read the raft paper and come away also knowing how group membership changes (wow) and log compaction works, in additional to the "basic" log replication consensus, that's huge.
edit: I'm sure it's still 'hard' to implement, like bringing a brand new replacement machine up from scratch (copying the full data set over, to warm him up, and then starting replication efficiently), i.e. there's still a lot of coding to do
If you can read the raft paper and come away also knowing how group membership changes (wow) and log compaction works, in additional to the "basic" log replication consensus, that's huge.
edit: I'm sure it's still 'hard' to implement, like bringing a brand new replacement machine up from scratch (copying the full data set over, to warm him up, and then starting replication efficiently), i.e. there's still a lot of coding to do