| In heterogeneous computing systems, computing nodes might be different whereas each computing node contains different system resources such as processors, graphics processing units, memories, networks, storage units, etc. These computing nodes and their internal resources have to collaborate well to provide required computing capacity.;Portability is always an issue in heterogeneous computing systems. Data generated on one machine might not be able to be used by others directly because of the incompatibility issues. Data type, endianness, size and padding situation are different in heterogeneous and even in homogeneous computing systems. Data marshaling procedure is indispensable, especially in open systems. In this thesis, a portable data exchange toolkit is proposed.;Meanwhile, scalability is another major issue for Grand-Challenge applications. How to utilize system resources efficiently is critical when problem size increases. Based on system resources' different features, workload should be scheduled properly among them. In this thesis, a novel GPU-based MD5-Blowfish encryption algorithm is designed and implemented to handle scaled data with some optimization features of NVIDIA Fermi architecture. |
