The efficacy of commutativity-based semantic locking in an industrial IP-PBX platform

Locking is the most common technique for concurrency control in databases. Traditionally, the notion of conflict for locking is based on read/write operations only, and if two operations are not in conflict they are said to be compatible. Following the advent of the object model, where object operations have richer semantics that can be exploited for greater concurrency, commutativity has been proposed instead to define the notion of conflict. The prospect of greater concurrency potential than traditional read/write locking has motivated many researchers to work on the topic of commutativity in the past, and today we are drawn into this research field for the same reasons. A literature survey of past work on commutativity has shown that up to this point research in the area is still mostly academic, with no evidence of application to real-world systems to see how well commutativity-based semantic locking (or commutativity locking for short) performs. For our research a real-world modern industrial software application programmed entirely in an object-oriented language together with the storage of its object-based data in an embedded database has been made available to us by our industry partners. The application is called IPES; and we were able to apply our work on commutativity to this application to evaluate the efficacy of commutativity locking. Our work on commutativity covers both theory and performance studies. We introduced a new way of commutativity static analysis, where we categorisedcommutative operation pairs into RR, DF and SC pairs. Doing so allows us to see the concurrency potential of a data set under object-level locking, field-level locking and commutativity locking. We designed and built a commutativity concurrency control mechanism called C3layer which implements commutativity locking and integrates into an object-oriented software application as a concurrency control layer between the application and its object database. The C3layer is also built to do field-level, object-level and serial locking; and a performance study comparing commutativity locking to all these other locking levels has been conducted. Tens of IPES transactions are selected and divided into three transaction mixes, and in each mix the sustained throughput of the system is measured for all four locking levels. The experiments are also repeated at various multiprogramming levels from 1 to 1,000. Our results showed that commutativity locking achieves better performance than serial locking, and on occasion it outperforms field-level locking and object-level locking as well. Finally, we share our experiences and raise issues about commutativity locking, as well as making some suggestions for future application designers considering a switch to commutativity locking