More observations on IDF (parallel programming)
By vtatkar on Aug 21, 2008
Day 2 (and for me, final day, I cant go back for Day 3) had a different flavor to it. I was glad to see how much work Intel is encouraging in the direction of parallel programming. Among the more interesting sessions was an Academic Roundtable on Multicore programming. With participants from Intel College, Tom Murphy of Contra Costa College (a local community college) and Dan Garcia from UC-Berkeley, the discussion centered around how little the academic community was doing to promote/teach parallel programming to incoming entrants. It is clear that future development will need designers and architects, not to mention implementation engineers, to clearly understand principles of parallel programming. A tidbit of interest was that the general survey of university curriculum indicates that most parallel programming teaching is limited to a 6-hour section as part of an OS course!
There was plenty of talk (at the roundtable) about OpenMP and MPI and automatic parallelism, but a couple of other observations particularly stuck out in my mind. One was the appreciation for lack of testing tools (ie. How I do know that the parallel program I have written is correct?) . The non-deterministic nature of this problem is perhaps the most troubling. Another was the observation by UCB faculty that they are thinking about this problem as a many-core problem, not just a multi-core problem (meaning, its not limited to 2, 4, 8 or 16 cores, but to 1000s of cores) and that scale brings its own unique challenges in thinking. A final point of interest was how little attention academia has generally provided over the virtues of programming for performance of which multi-core programming is but one aspect (I disagree with this sentiment but I could see several heads nodding in agreement and it is a significant viewpoint).
Tough topics to deal with... lets see how the academic community tackles these over the next few years. Otherwise, the burden of this training will fall squarely on the shoulders of employers. See here for more details .
On other topics, I noticed that interests were largely around the upcoming Nehalem chip architecture, power issues, visual computing challenges and opportunities and platform virtualization.
This was also a great opportunity to network with friends from both Intel and Sun!