Events & News
Colloquium
| Category | Lecture |
| Date | Thursday, October 2, 2008 |
| Time | 11:00 am |
| Location | GS 906 |
| Details | Light refreshments will be available at 10:45AM in the 9th floor atrium. |
| Speaker | Mohan Rajagopalan |
| Title | Senior Research Scientist |
| Affiliation | Intel, Programming Systems Lab |
Primitives for Forward-scalable Parallel Programming
The advent of mainstream parallel hardware brings to the limelight challenges such as performance, portability and programmability that have traditionally been associated with parallel programming. This talk focusses on experiences with a new approach to expressing and managing parallelism in a forward-scalable manner. A software prototype of the ideas includes a set of high-level primitives that are used to express different kinds of parallelism patterns and an underlying scheduling framework. Key distinguishing aspects of this work are: 1. the programming model is fully deterministic, 2. the primitives are composable as first class objects to enable automatic optimizations, and 3. finally, the use of non-traditional holistic approaches to reduce parallelism overheads. The general ideas appear to be promising, and this framework is currently being used in the back-end of a new language, Ct. Apart from preliminary performance numbers, this talk will also mention unusual optimizations that we've experimented with.
Biography
Mohan Rajagopalan is a Senior Research Scientist in Intel's Programming
Systems Lab. He leads a research effort focussing on new abstractions, and
optimizations for expressing parallelism. His current interests include
language and runtime technologies for emerging multi-core platforms, new
programming models such as for reusable and incremental computation, and
whole system optimization. Mohan received his Ph.D. degree from the
University of Arizona in 2006. He was the recipient of the 2005 IEEE/IFIP
Willam C. Carter Dissertation Award. His e-mail is mohan.rajagopalan at
intel.com.