Incorporating transaction semantics to reduce reprocessing overhead in replicated mobile data applications

Update anywhere-anytime-anyway transactional replication has unstable behavior as the workload scales up. To reduce this problem, a two-tier replication algorithm is proposed in (Gray et al., 1996) that allows mobile applications to propose tentative transactions that are later applied to a master copy. However it can suffer from heavy reprocessing overhead in many circumstances. We present the method of merging histories instead of reprocessing to reduce the overhead of two-tier replication. The basic idea is when a mobile node connects to the base nodes merging the tentative history into the base history so that substantial work of tentative transactions could be saved. As a result, a set of undesirable transactions (denoted B) have to be backed out to resolve the conflicts between the two histories. Desirable transactions that are affected directly or indirectly, by the transactions in B complicate the process of backing out B. We present a family of novel rewriting algorithms for the purpose of backing out B. By incorporating transaction semantics, our rewriting methods are strictly better at saving desirable tentative transactions than the traditional reads-from transitive-closure based approach. In most cases our rewriting methods are better at saving desirable tentative transactions than an approach which is based only on commutativity.