Method computer program product and system for maintaining replication

Abstract

An automated method of maintaining replication topology information is presented that allows system maintenance to occur with a minimum of system administrator intervention. For a replicated and distributed database across a network of computers, the network is divided or partitioned into sites that are islands of good network connectivity wherein communication between the site computers occurs using high bandwidth, low-latency remote procedure calls (RPCs). Further, intersite connections allow communication between servers of different sites using a low bandwidth, high-latency store and forward messaging scheme. Knowledge consistency checkers running independently on each individual server in the entire network assures that the servers within a site are automatically connected in a full-mesh replication topology within a site and that the sites themselves are connected via a spanning tree of intersite links. The knowledge consistency checker enforced replication topology ensures that replication changes will be transmitted over intersite links as efficiently as possible so that each change will be transmitted over each 'expensive' intersite link exactly once across a single direction in order to attain the minimum communication that must occur in order for all servers to be informed of such a change. A change arriving at a site replicates according to the intrasite replication mechanism.