The dbMark is a platform-independent benchmarking system dedicated to the evaluation of relational databases watermarking schemes. The system is implemented using Java technology and is therefore easily operational in any operating system that supports Java. The dbMark supports connectivity to MySQL databases using java database connectivity (JDBC) and recognizes database tables with numeric (integers and floating points) and non-numeric values (varchar). The system’s operation follows the well-known 3-phase operation of watermark benchmark platforms to embed a watermark in a digital asset; to perform attacks on the asset; and finally to run a watermark recovery process.
The dbMark prototype comes with a set of implemented watermarking schemes that can be directly used by scholars for training purposes, i.e. to test how to watermark a relational database table, or how to evaluate the performance of these watermarking schemes on a given database table, or how to build and load new watermarking schemes into the dbMark engine. The set contains the implementation of one method per watermarking scheme category: robust, fragile, and fingerprinting. Watermarking schemes can be implemented by users and loaded as plugins onto the dbMark corresponding cartridge. This pluggable architecture is also used for the implementation of the currently available schemes. The users are able to create their own watermarking schemes, with specific parameters and value types, and incorporate them as plugins into the system by using the Create Plugin component.
The dbMark prototype comes with a rich set of implemented attacks which can be performed against a watermarked database table. The users can configure an attack by choosing its type, a set of necessary parameters and, finally, by specifying the part of the database table that will be attacked, i.e. which attributes and/or tuples. The supported four types of attacks are the following: 1) The value modification attack, 2) The subset attack, 3) The sort/shuffle attack, 4) The additive attack.
Preserving the quality of the data is a critically important factor in watermarking and, as a result, many usability metrics have been devised to measure the amount of distortion raised by a watermarking scheme. Most of these metrics measure the distortion by comparing the data before and after watermarking it. The implemented usability metrics of the dbMark include: the mean change, the variance change and the mean squared error metrics, which can all work with numeric attributes (i.e. the metric will be loaded only if the database attribute that is to be watermarked is numeric) as well as the Levenshtein string similarity which can work with alphanumeric attributes, measuring the difference between string sequences. All of these metrics have been created in the form of plugins and they are parameterized to run during the watermark embedding phase.