MIT 6.824 Lec9 Chain Replication

Chain Replication

将replicated server组织成chain样式，如S1, S2, S3, S4。其中，S1为head node，S4为tail node。

Chain Replication处理写操作的机制如下：

• Client sends to head
• Forwarded down the chain, in order
• Each server overwrites old data with new data
• Tail responds to client

Chain Replication处理读操作的机制如下：

• Client sends to tail
• Tail responds (no other nodes involved)

CR的强一致性

当没有出现failure时，整个系统的reply都由tail node来完成，保证了强一致性。head node将requets按照一定的顺序向下传递，tail node按照该顺序对client进行回复。

当出现failure时，CR的恢复机制也比较简单：

• If head fails, successor takes over as head, no commited writes lost.
• If tail fails, predecessor takes over as tail, no writes lost.
• If intermediate fails, drop from chain, predecessor may need to re-send recent writes.

</