Electrical Engineering

Several parallel matrix multiplications developed since four decades based on decomposing the multiplied matrices into smaller size blocks, the blocks will be distributed among the processors to run matrix multiplication in shorter time than when only one processor will run the whole matrix multiplications. All parallel matrix multiplications algorithms suffer from three points: C The optimal size of the block of the decomposed matrices. C The optimal number and size of the exchanged messages between the processors. C Data dependency between the processors. And some algorithms suffer from fourth point, which is load balance especially with non-square matrix multiplications. In this paper we will introduce new parallel matrix multiplications to overcome the above four drawbacks, which will result in vast difference in performance in terms of time and load balance. For example, for a matrices multiplication of 5000×5000, it consumes 2812 seconds using cannon algorithm, while only 712 ...

In this paper we introduce new OWL class sets that can be obtained through and manipulated by Boolean operators. The need for new class sets come in order to process and manage complex OWL data, and to improve the expressive power of OWL language. A particular need for such sets arises in processing complex data involved in ontology for human disease which is being developed. The class sets introduced in this paper include the definition of minusOf, De Morgan's Law, and Auxiliary Identity properties.

In this thesis, a novel framework for paraUel processing is introduced. The main aim is to consider the modem processors architecture and to reduce the communication time among the processors of the paraUel environment. Several paraUel algorithms have been developed since more than four decades; aU of it takes the same mode of data decomposing and parallel processing. These algorithms suffer from the same drawbacks at different levels, which could be summarized that these algorithms consume too much time in communication among processors because of high data dependencies, on the other hand, communication time increases gradually as number of processors increases, also, as number of blocks of the decomposed data increases; sometime, communication time exceeds computation time in case of huge data to be parallel processed, which is the case of parallel matrix multiplication. On the other hand, all previous algorithms do not utilize the advances in the modem processors architecture. Matrices multiplication has been used as benchmark problem for aU parallel algorithms since it is one of the most fundamental numerical problem in science and engineering; starting by daily database transactions, meteorological forecasts, oceanography, astrophysics, fluid mechanics, nuclear engineering, chernical engineering, robotics and artificial intelligence, detection of petroleum and mineraIs, geological detection, medical research and the military, communication and telecommunication, analyzing DNA material, Simulating earthquakes, data mining and image processing. In this thesis, new parallel matrix multiplication algorithm has been developed under the novel framework which implies generating independent tasks among processors, to reduce the communication time among processors to zero and to utilize the modem processors architecture in term of the availability of the cache mem. The new algorithm utilized 97% of processing power in place, against maximum of 25% of processing power for previous algorithms. On the hand, new data decomposition technique has been developed for the problem where generating independent tasks is impossible, like solving Laplace equation, to reduce the communication cost. The new decomposition technique utilized 55% of processing power in place, against maximum of 30% of processing power for 2 Dimensions decomposition technique. v Foreword I dedicate this work frrst of aIl to my parents Hussein and Inaam, who partied the nights on my upbringing. I dedicate this work to my life partner and so my PhD partner .... Rana I foreword my work to the scientist of Math, to AI-K.hwarizmï, Abü Ja'far Muhammad Ibn Müsa, Pythagoras (IIu9U'yopaç), Isaac Newton, Archimedes, René Descartes, and Alan Mathison Turing, father of computer science. I foreword this work to my daughter Leen, and my sons Hussein, Yazan, and Muhammad. I would acknowledge proudly my supervisor Prof. Adel Omar Dahmane, for his unlimited support through my PhD march, and I would thank aIl who supported me by aIl means.