http://book.mixu.net/distsys/replication.html
Here are some of the key characteristics of each of the algorithms:
Primary/Backup
- Single, static master
- Replicated log, slaves are not involved in executing operations
- No bounds on replication delay
- Not partition tolerant
- Manual/ad-hoc failover, not fault tolerant, “hot backup”
2PC
- Unanimous vote: commit or abort
- Static master
- 2PC cannot survive simultaneous failure of the coordinator and a node during a commit
- Not partition tolerant, tail latency sensitive
Paxos
- Majority vote
- Dynamic master
- Robust to n/2-1 simultaneous failures as part of protocol
- Less sensitive to tail latency
Understanding replication in databases and distributed systems
https://ieeexplore.ieee.org/document/840959
Replication Techniques for Availability
https://link.springer.com/chapter/10.1007/978-3-642-11294-2_2