AllConcur: Leaderless Concurrent Atomic Broadcast(arxiv.org)
arxiv.org
AllConcur: Leaderless Concurrent Atomic Broadcast
http://arxiv.org/abs/1608.05866
6 comments
Not really fair to compare to Paxos if the network assumption is different.
Sounds like Paxos could be implemented on top of UDP whereas AllConcur requires TCP?
Duplicated and lost messages will happen in practice - TCP attempts to build reliable communication on an unreliable medium. Not clear from the paper if this matters.
I think the larger misunderstanding of the paper is the requirement that paxos have a leader - lamport's synod algorithm does not require a leader, the leader is an optimization!
I think the larger misunderstanding of the paper is the requirement that paxos have a leader - lamport's synod algorithm does not require a leader, the leader is an optimization!
Without a long-lived leader, the performance would be so bad that I doubt there is a single production implementation that doesn't use that optimization.
How does this differ from RAFT?
"We assume a model of reliable communication—messages cannot be lost (only delayed). This is a reasonable assumption if we consider a reliable protocol, such as TCP."
Paxos network assumptions:
"Messages can take arbitrarily long to be delivered, can be duplicated, and can be lost, but they are not corrupted." - http://research.microsoft.com/en-us/um/people/lamport/pubs/p...