Tesis

My Ph.D. Thesis, titled "Supporting general data structures and execution models in runtime environments", covers topics related with tools for parallel programming and parallel execution. It studies the support these tools give to handle dense and sparse data structures, and the execution of tasks based on dataflow computations.

Supervisor

• Dr. Arturo Gonzalez Escribano

Graduation committee

• Dr. Clemens Grelck (University of Amsterdam, Netherlands) -- Committee president
• Dr. Luis Alberto Marqués Cuesta (University of Valladolid, Spain) -- Committee secretary
• Dra. Dora Blanco Heras (University of Santiago de Compostela, Spain)
• Dra. Maria Angeles González Navarro (University of Málaga, Spain)
• Dr. Francisco Javier García Blas (University Carlos III de Madrid, Spain)

Grade

Sobresaliente cum laude

Abstract

Parallel computing systems have become increasingly popular over the past decades in both high performance and mainstream computing. The reason is that there has been a huge increase in the capabilities of parallel platforms. To take advantage of these platforms, we require new programming tools. First, models with high-level abstractions to represent appropriately parallel algorithms. Second, the programming frameworks that implement these models need complete runtime systems that offer different parallel paradigms to the framework programmers. This way, it is possible to build programs that can be efficiently executed in different platforms.

There are different areas to study in order to develop a complete runtime system for a parallel framework. This Ph.D. Thesis addresses two common problems: The unified support for dense and sparse data, and the integration of data-mapping and data-flow parallelism.

In the first place, a generic parallel framework requires to integrate dense and sparse data management using a common interface whenever possible. We propose a solution that decouples data representation, partitioning, and layout from the algorithmic and from the parallel strategy decisions of the programmer.

In the second place, such parallel framework should allow to program dynamic and dataflow applications, which is a challenging task with current available tools. In this Thesis, we introduce a new programming model based on the dataflow paradigm, where different activities can be arbitrarily linked, forming generic but structured networks that represent the overall computation.

Download

A PDF version of the Thesis and another PDF with the presentation slides can be found in the following links: