An analysis of borrowing policies for escrow transactions in a replicated data environment

Three issues in the context of escrow-based methods, are addressed. It is shown that full serializability can be guaranteed for escrow transactions by means of repeat semantics, that is, a transaction which fails is rerun in a special mode. The author describes a crash recovery protocol consistent with the concurrency control scheme used. An analysis is made, by means of a simulation model, of several different borrowing policies for escrow transactions and the best one is identified.<
>