Massive parallel processing for graph pattern matching on EMF models

OData support
Dr. Hegedüs Ábel
Department of Measurement and Information Systems

In this document a system is described that is capable of massively parallel graph pattern matching over EMF models. This system is able to generically process EMF metamodels and models, and then match graph patterns over them, utilizing the GPU's highly parallel architecture. When patterns are matched, the same operations are processed for all of the entities of a given type in the model. Using a GPGPU approach to process these patterns in a parallel way can increase the performance of pattern-matching.

Graph pattern-matching has many applications and in some cases models comprise millions of elements. It is a computationally intensive task to match complex patterns over these large models.

EMF-IncQuery is a framework for defining model queries over EMF models. Its advantage is that queries can be provided declaratively, with no manual coding required. The system is planned and implemented as an extension for EMF-IncQuery. This system is able to take EMF-IncQuery's pattern models and without manual coding process them using the GPU. The results are also transparent, which means that the caller should not be able to distinguish between the results of EMF-IncQuery and the system described in this paper.

The system designed could help utilizing the graphics processing units present in most of the computers today for pattern-matching purposes. It also presents a possible way to handle very large models and complex patterns.


Please sign in to download the files of this thesis.