Charla: High Performance Distributed Computing Systems

19 de abril, 2013

El miércoles 24 de abril, a las 11:30, el profesor visitante Hermes Senger impartirá la charla “High Performance Distributed Computing Systems“. La actividad se realizará en la sala de postgrado del Departamento de Ingeniería Informática.

El Dr. Senger es profesor asistente de Ciencias de la Computación en la Universidad Federal de San Carlos en Brasil, desde febrero de 2009. Sus áreas de investigación incluyen la computación de alto rendimiento, aplicaciones y sistemas computacionales paralelos y distribuidos, planificación de recursos y análisis de escalabilidad. La charla busca presentar formas novedosas y útiles para diseñadores de aplicaciones y arquitecturas, para producir sistemas computacionales más escalables.

El profesor Senger se encuentra en Chile entre el 20 y 27 de abril en el contexto de un proyecto STIC-AMSUD de cooperación internacional entre Francia, Brazil, Argentina y Chile, coordinado por el académico del Departamento de Ingeniería Informática, Dr. Mauricio Marín.

 

Abstract

High performance distributed computing systems (HPDC) such as clusters, grids, and computational clouds are becoming ubiquitous. Within organizations, systems such as MapReduce are being employed to process several dozens of petabytes of data produced every day by commercial transactions, stock exchange, social networks, and other data sources in order to obtain competitive advantages. Within the academia, the combination of high performance computing systems and special devices (e.g., particle accelerators, DNA sequencers, radio telescope arrays, etc) interconnected by high bandwidth networks are changing the way science is made.
Nowadays, most powerful parallel systems are composed of hundreds of thousands of processors working simultaneously. Currently, a big challenge is the development of applications that can be scalable to this level, taking full advantage of the architecture to support the appropriate concurrency requirements. In this talk, some scalability limits of important parallel applications (e.g., Bag-of-Tasks, and MapReduce) will be discussed. From our discussion, we hope to provide useful insights for application and architecture designers to produce more scalable computational systems.