English | 中文 | Русский | Français

2,856 Posts served

8,606 Conversations started

UPCRC Illinois Summit - 12 FEB 2009

By Clay Breshears (Intel) (133 posts) on February 24, 2009 at 3:20 pm

I was able to attend the Universal Parallel Computing Research Center (UPCRC) Illinois Summit meeting on 12 FEB 2009. This was a progress report of various research projects that are going on with faculty and students at the University of Illinois at Urbana-Champaign. It's been about a year since the UPCRC program was initiated with grant money from Intel and Microsoft.  The faculty have gotten projects started and students have been brought in to lend a hand.

To keep this post short, I just want to highlight some of these projects that caught my attention.  For more information on what UPCRC Illinois is doing, you should check out the project webpage at http://www.upcrc.illinois.edu/.  This page features links to upcoming events, the various lecture series sponsored by UIUC (some of which can be viewed via streaming video), and the research that is going on here in Illinois.

To get things started, the statement "Parallel Programming (as practiced today) is Hard" was put forth.  I think this sets the tone for the whole UPCRC effort at UIUC.  Their "catchphrase," that sums up the mission of Illinois's UPCRC research, is "Making Parallelism Easy".

The first project was for Video Event Detection, which seeks to be able to have computers scan video footage looking for a specific event.  How many time have I seen the cops in "Law & Order" watch security camera video looking to see if the white van drives by?  If they could have computers do the initial looking, that would probably solve cases faster. After that was a report on 3-D propagation that uses a depth camera in conjunction with fixed video images of a scene to better render images of the scene as visualized by a virtual camera. The Tele-Immersion project, with its shared virtual environment reminded me of the boardroom scene from one of the "Resident Evil" movies.  The members could be at different places, but they all meet at the same table and are able to collaborate, at least this is what it looks like to the participants.  An effort is underway, with cooperation of Berkeley UPCRC researchers, to put together an encyclopedia of parallel programming patterns.  Dr. Ralph Johnson, one of the "Gang of Four" authors, is leading the effort at Illinois.  There will be a ParaPLoP (Parallel Pattern Language of Programs) workshop in Santa Cruz in early June.  For more information and call for papers (ends 18 APR 09) see ParaPLop 2009.

The Deterministic Parallel Java language project is trying to demonstrate that a programming language for parallel execution can be deterministic by default.  Any non-determinism that the programmer requires must be specified explicitly.  This supports the conclusions of the some of the presenters from the previous day's panel on Determinism vs. Non-determinism.  The final talk of the day concerned what hardware design changes would be needed to support disciplined parallel programming.  Before that could be realized, the semantics for a concurrency model that banished data races would be needed. 

Another hardware project presented the Bulk Multicore Architecture.  The idea for this was to have the hardware execute chunks of dynamic instructions, say 2000.  Chunks execute atomically and in isolation to achieve deterministic behavior.  The chunk executions would be considered as transactions that could be rolled back and reexecuted if there was a detected conflict.  The "hook" behind the 'transactions all the time' execution model is that this would be unknown to the software.  Thus, programmers write up their parallel algorithms, and the architecture would automatically handle the bulk transactions.  There is a forthcoming article to be published in Communications of the ACM.

There were quite a few more projects reported.  I was really impressed by the range of projects that the UIUC faculty were covering and the progress that's been made in the first year.  It will be interesting to see how the current researchers wind up and what spin-off and new investigations those results lead to.

UPDATE (25 FEB 09): Video of the sessions with presentations for this day are available at http://www.upcrc.illinois.edu/workshops/summit_feb2009/agenda.html

Categories: Academic, Parallel Programming

Comments (0)

Trackbacks (0)

Leave a comment