DATA REPLICATION USING EPHEMERAL TREE STRUCTURES

摘要

The disclosure is directed to replicating data across multiple computer nodes (“nodes”) in a distributed computing system. The data is replicated to a significantly large number of servers in two modes—a pull mode and a push mode. In the push mode, data replication is initiated by a publisher node, which replicates the data to the nodes. In the pull mode, data replication is initiated by a node which pulls the data from one of the nodes. The nodes are deployed as an ephemeral tree structure (“tree”) such that data can flow efficiently between nodes of different hierarchical levels of a communication network, e.g., a region, a datacenter, a cluster, and a rack. Data is transmitted across the hierarchical levels from a leader node of one level to a leader node of another level, and from the leader node to the other nodes within the level.