Scalable Design-Space Exploration

OData support
Supervisor:
Dr. Horváth Ákos
Department of Measurement and Information Systems

The goal of Model Driven Engineering (MDE) is to achieve a system design starting

from a high level of abstraction and then refining it gradually into a specific

realization of a solution. As a result of this approach the initial model of the design

problem is not specific enough to automatically generate a solution, rather it spans

a design space of models possibly containing a large number of solutions.

Design Space Exploration (DSE) is a search based methodology that finds an

appropriate solution within such a design space that is sufficiently good with regards

to multiple criteria. These criteria however in the case of MDE are more often

than not described by complex structural requirements over the model like network

connectedness or model interdependency, in which cases the otherwise widely used

methods of logical programming or SAT solvers are either hard to apply or do not

scale sufficiently well.

In response to the drawbacks of traditional approaches several research projects

started trying to define unified frameworks based on the MDE approach that are

specifically designed to deal with such requirements. VIATRA-DSE is a framework

that achieves this by defining graph patterns and transformation rules to be used

within a graph pattern matching engine. The used engine is called EMF-IncQuery

and it provides highly scalable incremental pattern matching for outstanding performance.

The VIATRA-DSE framework has two important drawbacks: (i) it can

only operate on models defined in it’s own modeling environment and (ii) due to it’s

monolithic architecture it cannot scale sufficiently well to very large problems.

The scope of this Master’s Thesis is to improve the scalability capabilities of

the previously developed VIATRA-DSE framework while utilizing as much of the

previous research as possible in order to reach a level of performance that enables

the framework to be used in industrial scale design problems. In order to achieve this

goal, I have (i) used most of the techniques already in the VIATRA-DSE framework,

(ii) improved upon the system by extending it with the ability to communicate the

design problem data over the network and made it possible to (iii) do computations

on multiple computers simultaneously, greatly improving the overall performance.

Downloads

Please sign in to download the files of this thesis.