Multicore problem
in [ADA]
|
Prev: Bitbucket.org now supports Ada and Modula-2 syntax highlighting
Next: Q: Stopping a task running a long simulation
From: Anatoly Chernyshev on 10 Mar 2010 02:19 Thanks a lot to all responded. What is clear now is that the problem is rather complicated. I guess, I'd better be off with distributing the calculation over several PCs, with one task on each.
From: Charmed Snark on 10 Mar 2010 13:00 Anatoly Chernyshev expounded in news:095dc775-e214-4ccd-bf26-45ab27b3b277 @s36g2000prh.googlegroups.com: > Thanks a lot to all responded. What is clear now is that the problem > is rather complicated. I guess, I'd better be off with distributing > the calculation over several PCs, with one task on each. I was wondering about these issues in connection with pthreads. I have a basic interpreter project (https://sourceforge.net/projects/bdbbasic) where I was considering implementing matrix operations in multiple threads. The net wisdom suggested that the thread startup cost was around 100usec as a general rule. However, that appears to be only the beginning of important considerations. ;-) From the related post here, it seems that the matrix might need to be huge to be of benefit. But even then, as the cores and cache interact, this might not improve performance. Interesting stuff indeed. Then when you add this to multiple platforms, a project designer really gets his hands dirty.. Warren
From: tmoran on 10 Mar 2010 14:24
>Thanks a lot to all responded. What is clear now is that the problem >is rather complicated. In the early days of virtual memory, one of our users (at U of Wisconsin) decided to make use of the large memory by declaring large arrays, which his program accessed by running down columns. Unfortunately, the compiler allocated them by rows, so his program's thrashing brought the whole system to a crawl. I imagine there will have to be a similar learning process for using multicores well. |