1. UKM Learning and Research Repository
2. PHD Thesis / Tesis PHD
3. Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Title:	Improvements of performance for parallel simulation of membrane computing with new methods and algorithms
Authors:	Maroosi Ali (P64621)
Supervisor:	Ravie Chandren Muniyandi, Dr.
Keywords:	Membrane computing Molecular computers
Issue Date:	Feb-2015
Description:	Membrane computing obtained the computing models from the architecture and functionality of the living cell. It works within parallel and distributed computing models, in which multisets of objects located in the compartments enclosed by membranes, and develop according to evolving rules. Membrane systems exhibit two levels of parallelism: the rules are applied in parallel (region level parallelism) and the membranes evolve concurrently (system level parallelism). Appropriate criteria to design membrane systems to have enough parallelism, and the methods and algorithms to exploit the parallelism are important factors in the performance of parallel simulation of membrane computing. The previous approaches have not used appropriate methods and algorithms to exploit the parallelism in the membrane computing. The thesis introduces new criteria, methods and algorithms to address shortcoming of the previous approaches. Firstly, the thesis introduces new criteria to consider the types of rules beside the number of applied rules per step to improve the way to design a membrane system. The method of applying fewer number of communication rules per step between membranes that perform same tasks is the better way of designing the parallel simulation of a membrane system. Secondly, the thesis introduces methods and algorithms to improve simulation of the membrane systems on the graphics processing units (GPUs) which are cost effective with many cores. In this study membrane systems were represented as matrices to make effective allocations to the threads and the thread blocks, and to increase the multiprocessor occupancy on the GPU. New weighted networks to determine the rate of communications between the objects and between the membranes were introduced. A clustering algorithm based on those networks were also introduced to classify the dependent objects and membranes to the same groups, and assigns them to the same threads and threads blocks as much as possible to reduce the communications between the threads and the thread blocks and to increase the performance. Finally, the thesis demonstrates some of the introduced methods to enhance the membrane inspired intelligent algorithms. Unlike the previous approaches that applied communication task in each step, to improve the parallel execution, membrane inspired algorithms to exchange the information between the membranes with low rate are introduced. Simulation of the designed membrane systems with the new criteria could achieve the speedup up to 1.8 times over multi-core processors compared to sequential approach, while for conventional membrane system design, it is 100 times slower than sequential approach. The proposed methods and algorithms on the NVIDIA Geforce 680 GPU demonstrate high performance compared to previous studies. In some cases the speedup was up to 82 times for the proposed approaches, while for previous studies speedup is 8.2 times. The introduced membrane inspired intelligent algorithms could achieve better solution than previous approaches with the same simulation time for the optimization problems.,Ph.D
Pages:	246
Call Number:	QA76.887 .M365 2015 3
Publisher:	UKM, Bangi
Appears in Collections:	Faculty of Information Science and Technology / Fakulti Teknologi dan Sains Maklumat

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.