Parallel Statistical Computing: Are We Embracing the Scalable Concurrency Revolution?

• Activity Number: 189
• Type: Invited
• Date/Time: Monday, August 2, 2010 : 10:30 AM to 12:20 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): George Ostrouchov*+
• Companies: Oak Ridge National Laboratory
• Keywords: parallel computing ; GPU ; software engineering ; algorithms ; distributed computing

Abstract

Increasing processor clock rates enabled increases in statistical computing complexity and enabled the processing of larger data sets for many years. Statistical software would just run faster on a new chip with a faster clock rate. This free ride is now over until we fully embrace the concurrency revolution and the development of scalable parallel algorithms. The last five years have seen stagnant clock rates and more processors. Why? There are fundamental reasons for this change and the future is full of multiple cores and specialized co-processors. Serial algorithms get little benefit from a second processor. Scalable parallel algorithms now get a free ride. Where will they come from and who should provide them? This poster presentation will attempt to answer some of these questions. The new computing landscape presents both challenges and opportunities for statistics.