Parallel Approaches to Shortest-Path Problems for Multilevel Heterogeneous Computing
[Download pdf] [Slides]
Distinction Excellent Cum Laude
ABSTRACT
Many graph algorithms have given solution to problems of finding shortest paths between nodes of a network. These problems are considered among the fundamental combinatorial optimization problems. During the last years, the interest of the scientific community to this kind of problems has significantly increased not only due to its wide-applicability, but also thanks to the currently popular and efficient parallel computing. Additionally, the advent of new parallel programming models together with modern powerful hardware accelerators, such as the Graphics Processing Units, opens the possibility to study new and more efficient parallel approaches to exploit these specific architectures. Furthermore, the emerging of heterogeneous parallel computing combining these powerful hardware accelerators with the classical and increasingly powerful CPUs, provides a perfect environment to face the most costly shortest-path problems in the context of High Performance Computing.
The main goals of this Ph.D. Thesis are to develop new GPU-based approaches to the shortest path problem, and design solutions where both sequential and parallel algorithms are deployed concurrently in heterogeneous environments. Moreover, we introduce a multilayer abstract model that helps the programmer to easily obtain flexible and portable programs that automatically detect at run-time the available computational resources and exploit hybrid clusters with heterogeneous devices.
ADVISORS
- Dr. Diego R. Llanos Ferraris, and
- Dr. Arturo González Escribano.
COMMITEE MEMBERS
- [President] Prof. Enrique Salvador Quitana Ortí -- Universidad Jaume I,
- [Member] Dra. Ana Lucía Varbanescu -- University of Amsterdam,
- [Member] Dra. Rosa Filgueira Vicente -- University of Edinburgh,
- [Member] Dr. Gracia Ester Martín Garzón -- Universidad de Almería,
- [Secretary] Dr. Mario Martínez Zarzuela -- Universidad de Valladolid.