1. Introduction
- many internet and cloud computing applications have scaling requirements that exceed the capabilities of enterprise databases
- solve: use sharding on a cluster of commodity servers
- disadvantages: To ensure linear scaling, the scope of a transaction is usually limited to a single node.
lead to maintenance and load balancing headaches.
- many internet and cloud computing applications also need to be continuously available.
- solve: One solution is to use synchronous master-slave replication on pairs of nodes to provide this capability.
- disadvantages: 1.1
1.1 Limitations of Master-Slave Replication and the Case for Paxos
- 2-way synchronous replication and 3-way synchronous replication's disadvantages.
- Paxos solves the general problem of reaching consensus on the state of 2F + 1 replicas while tolerating up to F failures.
1.2 Strong vs. Eventual Consistency
1.3 Spinnaker
- In CAP terminology, Spinnaker is an example of a CA system.Explanation of CAP theorem
- It is designed for a single datacenter and assumes that a different replication strategy is used for cross-datacenter fault tolerance (presumably an asynchronous one).
