Application of Partition Serializability in Software Engineering Databases

It is widely recognized that classical transaction concepts do not su ciently support software engineering applications. Traditional transactions are expected to have short duration while software engineering transactions may last for a long time, and concurrently access a large number of objects. If software engineering transactions are executed in a serializable manner, they will block each other for long periods of time. This paper shows how parameterized Partition Serializability (PSR) can be applied to software engineering databases (SEDBs) to enhance transaction concurrency. PSR is based on partitions of object spaces. Di erent partitions of the same object space can be obtained. Although not every partition of an SEDB object space can ensure complete consistency, a good partition ensures that inconsistency will occur rarely and be in fairly controllable forms that can be handled manually (say) at transaction termination. The paper discusses approaches for applying PSR in SEDBs, and demonstrates its applicability by giving examples of SEDB object spaces, object space partitions, and useful PSR schedules that are not serializable. We argue that the PSR approach is more realistic than approaches relying on the user only to resolve con icts, as well as approaches based on correctness criteria that can be used to automatically resolve con icts.