Autonomous and decentralized replication in the Pangaea planetary-scale file service

Pangaea is a planetary-scale file service that supports sharing of frequently written files as well as read-only files. Pangaea uses massive replication to achieve low access latency and high availability. This paper focuses on a key feature of Pangaea— randomized, optimistic protocols for managing many replicas efficiently for billions of files replicated on hundreds of servers. Replica membership is maintained as a sparse but strongly con- nected graph, per file. Updates to the membership and contents are propagated along the graph edges. The feasibility of Pan- gaea is evaluated using a prototype and a simulation.