Prev: PEEEEEEP
Next: Texture units as a general function
From: =?ISO-8859-1?Q?Niels_J=F8rgen_Kruse?= on 2 Jan 2010 12:20 Robert Myers <rbmyersusa(a)gmail.com> wrote: > Some engineering applications make good use of large-scale > computation. One aerodynamicist I talked to who used CFD as a black > box said he was convinced there was a bug in a program that was widely > relied upon for aerodynamics. Even there, the successes may be > substantially delusional/luck. In the benchmark <http://www.r-goetz.de/RGBench/bench.shtml>, i noticed a bug in mci_bench.c (a Monte-Carlo-simulation of a lattice model of a simple fluid). (I tried to contact the author in 2006, but failed.) The bug is in the line "while(dx==dy==dz==0)". I think it is intended to weed out zero displacements. I trust readers can spot the problem. Just en example of a code that appears to have been used for serious purposes. -- Mvh./Regards, Niels J�rgen Kruse, Vanl�se, Denmark
From: Del Cecchi on 2 Jan 2010 21:15 "Robert Myers" <rbmyersusa(a)gmail.com> wrote in message news:ac24ed81-04ff-46fc-b067-8f38ca97b543(a)z41g2000yqz.googlegroups.com... On Jan 2, 4:32 am, Mayan Moudgill <ma...(a)bestweb.net> wrote: > Robert Myers wrote: > > > The scientists I know generally want to speed things up because > > they > > are in a hurry. > > > > The question is: is it better to do a bit less physics and/or let > > the > > machine run longer, or is it better to use up expensive scientist/ > > scientific programmer time and, at the same time, make the code > > opaque > > and not easily transportable? > > > > If we can't do "unbounded" ("scalable") parallelism, then there is > > an > > end of the road as far as some kinds of science are concerned, and > > we > > may already be close to it or even there in terms of massive > > parallelism (geophysical fluid dynamics would be an example). The > > notion that current solutions "scale" is pure bureaucratic fraud. > > Manufacturers who want to keep selling more of the same (do you > > know > > any?) cooperate in this fraud, since the important thing is what > > the > > customer thinks. > > > > If your problems can be solved by simply increasing the number of > machines, why not go with Beowulf clusters or @Home style > parallelism? > They are cheapa and easy to put together. > I don't have a clue about @Home style parallelism. When I see that the approach has produced real science, I might get interested. My suspicion is that the approach is very inefficient as to the cost of energy and manpower. The approach would really only make sense if hardware were expensive and energy and sysadmin time were free. If those were ever the case, they aren't now. For mid-scale problems, clusters built around fairly beefy nodes (a 2 socket 12-core might be a sweet spot in terms of cost for now, a 4 socket 24-core machine might be even better in terms of dealing with interconnect), might be fairly attractive compared to just about anything you can buy commercially. If the warehouse-filling clusters of not so long ago were interesting, then what you can build today should be even more interesting. Since I've argued that super-sized computers seem to me to be of questionable value, maybe that's all the vast bulk of science really needs. If I really need a computer with respectable bi-section bandwidth, I can skip waiting for a gigantic machine that runs at 5% efficiency (or worse) and learn to live with whatever I can build myself. > If your problem can't be solved with those approaches, then I > suspect > that going to a different language (or approach, or whatever) is not > going to be a viable alternative. I agree. The only thing a better approach would deliver is more ease of use, but that's more important than the cost of the hardware. Well, maybe I agee. The typical 5% sustained performance of the big machines would seem to leave a lot of room for improvement that could conceivably be addressed by slightly less clueless approaches to massively parallel software. Robert. -----------------------end quote---------------- In the case of @home, both the energy and hardware are "free" since they are donated by folks who believe that the energy would be used anyway to run the hardware that would otherwise be sitting idle. del
From: Robert Myers on 2 Jan 2010 22:13 On Jan 2, 9:15 pm, "Del Cecchi" <delcec...(a)gmail.com> wrote: > > In the case of @home, both the energy and hardware are "free" since > they are donated by folks who believe that the energy would be used > anyway to run the hardware that would otherwise be sitting idle. > That's no longer true. If I use all four cores of my i7, it uses at least twice as much power as if it sits "idle" without hibernating. I assume that all new hardware is going to work similarly. While the project may not be footing the bill for the extra energy, someone is, and the associated greenhouse gases are most definitely billowing into the atmosphere in a way they would not be if the @Home calculation weren't running. It's conceivable that I could find a combination of @Home settings and Computer Power Management setting that would result in negligible extra power, but how many will do that? In any case, the idea that the computers would be running, anyway, no longer works as cleanly as it once did. Al Gore would be extremely displeased, and we surely wouldn't want that. Robert.
From: Andrew Reilly on 2 Jan 2010 23:46 On Sat, 02 Jan 2010 19:13:57 -0800, Robert Myers wrote: > It's conceivable that I could find a > combination of @Home settings and Computer Power Management setting that > would result in negligible extra power, but how many will do that? I'm fairly confident that Mayan's use of @Home as a tag was more along the lines of "computations that are known to require lots of grinding on small chunks of data, and so therefore trivially parallelizable with arbitrary network geometry", rather than the "free cycles" aspect. I have occasionally wondered just how many "interesting" problems fit that model. Certainly some classes of search or optimization, but many that use real or interesting data? Part of this discussion that I find particularly interesting is the distinction between "arbitrarily scalable" code (if such exists) and code that potentially has a largish but problem-specific "right" number of nodes. I'm fairly sure that most of the problems that I find interesting are in the latter camp: past a certain point structuring the code to make use of more processors would require at least a completely different approach to decomposition, if it is possible at all. Cheers, -- Andrew
From: EricP on 2 Jan 2010 23:57
Andrew Reilly wrote: > > I have occasionally wondered just how many "interesting" problems fit > that model. Certainly some classes of search or optimization, but many > that use real or interesting data? Protein folding distributed computing to understand protein folding, misfolding, and related diseases http://folding.stanford.edu/ Eric |