A distributed transaction processing protocol based on majority consensus

Abstract

An n-failure-resilient protocol for distributed transaction processing regarding site crash, network partitioning, and local failures like abort of subtransactions is specified. The three-phased protocol contains concurrency control, a mechanism enforcing mutual consistency, and commitment based on majority consensus w.r.t. multiple copies of entities. To cope with transaction manager failures, backup processes for the manager are used. Data redundancy given by multiple copies in a partially redundant database is utilized to improve the resiliency of the protocol, i.e. to improve especially the successful commitment of update transactions processed by the protocol. The protocol is n-failure-resilient if the least redundant entity to be accessed by a transaction is mapped onto at least 2n+1 file copies in the distributed database.