Master Thesis - Past Projects - Abstract

Data Versioning in a Real-Time Data Repository

ID: LiTH-IDA-EX-04/070-SE

Most engines in modern cars are controlled by a computer called Engine Control Unit (ECU). The software running on the ECU can be divided into two parts: functionality for controlling the engine and functionality for diagnosing faults in the engine. In order to deal with increasing complexity of the software a database has been implemented as part of a previous project. As the diagnosis is not as time-critical as the functionality for controlling the engine, it is run with low priority. This makes it hard to guarantee relative consistency for the data used by the diagnosis. Previous work has shown that multiple versions of data can help increase both temporal consistency of data and performance of the system. This thesis presents implementation and performance evaluation of two concurrency control algorithms, optimistic concurrency control (OCC) and multi-version timestamp ordering validity bounds (MVTO-VB). MVTO-VB utilizes multiple versions of data in order to guarantee relative consistency. The performance evaluation compares the algorithms in conjunction with two different on-demand updating algorithms: On-Demand Depth First Traversal (ODDFT) and On-Demand Top-Bottom (ODTB). It shows that in conjunction with ODDFT, MVTO-VB obtains the best performance among the algorithms. In conjunction with ODTB, a variant of OCC obtains the best performance and MVTO-VB the worst.

Keywords: real-time databases, concurrency control, relative consistency

Author(s): Hugo Hallqvist

Contact: Jörgen Hansson