Transactional concurrency control for resource constrained applications

Abstract

Transactions have long been used as a mechanism for ensuring the consistency of databases. Databases, and associated transactional approaches, have always been an active area of research as different application domains and computing architectures have placed ever more elaborate requirements on shared data access. As transactions typically provide consistency at the expense of timeliness (abort/retry) and resource (duplicate shared data and locking), there has been substantial efforts to limit these two aspects of transactions while still satisfying application requirements. In environments where clients are geographically distant from a database the consistency/performance trade-off becomes acute as any retrieval of data over a network is not only expensive, but relatively slow compared to co-located client/database systems. Furthermore, for battery powered clients the increased overhead of transactions can also be viewed as a significant power overhead. However, for all their drawbacks transactions do provide the data consistency that is a requirement for many application types. In this Thesis we explore the solution space related to timely transactional systems for remote clients and centralised databases with a focus on providing a solution, that, when compared to other's work in this domain: (a) maintains consistency; (b) lowers latency; (c) improves throughput. To achieve this we revisit a technique first developed to decrease disk access times via local caching of state (for aborted transactions) to tackle the problems prevalent in real-time databases. We demonstrate that such a technique (rerun) allows a significant change in the typical structure of a transaction (one never before considered, even in rerun systems). Such a change itself brings significant performance success not only in the traditional rerun local database solution space, but also in the distributed solution space. A byproduct of our improvements also, one can argue, brings about a "greener" solution as less time coupled with improved throughput affords improved battery life for mobile devices.