Each replica maintains a local copy of the log. The
Paxos algorithm is run repeatedly as required to ensure that all replicas have identical sequences of entries
in their local logs. Replicas communicate with each other through a Paxos-specific protocol.
The next layer is a fault-tolerant replicated database which includes a local copy of the database at each
replica. The database consists of a local snapshot and a replay-log of database operations. New database
operations are submitted to the replicated log. When a database operation appears at a replica, it is applied
on that replica’s local database copy.
Finally, Chubby uses the fault-tolerant database to store its state. Chubby clients communicate with a
single Chubby replica through a Chubby-specific protocol.