Abstract
Parallel computing is anticipated to become a mainstream practice in most areas of computing. On the verge of the end of Moore's law, most processor manufacturers are introducing multi/many-core chips. While potentially improving system throughput and power consumption, each application cannot improve its performance unless it is parallelized. Unfortunately, we are currently in a situation where we have parallel architectures while not having parallel computing for most applications.
Parallelization of applications has been practiced intensively for scientific computing and specialized server applications. The development of parallel applications has the reputation of being difficult and error prone. Inheriting parallel computing practices in general-purpose computing domain is further challenged by the additional attributes of application interactivity, the fine-granularity of the work, and the less portability of parallel programming paradigms.
In this talk, we will address the new challenges that obstruct the transition to the ubiquity of parallel computing. We will present our experiences in optimizing parallel applications and enhancing the performance of parallel architectures both for general-purpose computing and embedded systems.
Finally, we will outline research areas we are currently focusing on to pave the way for making parallel computing mainstream.
Biography
The speaker is currently a visiting researcher in INRIA Institut National de Recherche en Informatique et Automatique, France. He got his PhD from North Carolina State University in 2003. He conducted research in different aspects of parallel architecture and computing. Specifically, his research efforts include speculative execution on shared memory CMP-based systems, power-aware scheduling for multiprocessor systems and design space exploration for multiprocessor systems. He recently researched code optimization for high performance computing and the use of GPUs and Cell BE as accelerators. Currently, he is interested in the transition era of parallel computing to commodity applications. For more details about the speaker consult his homepage: http://www.irisa.fr/caps/people/kibrahim