Hello Simon, Tuesday, April 17, 2007, 11:51:07 AM, you wrote:
The parallel GC code is currently sitting dormant waiting for me to go back to it and figure out why it isn't performing very well. Basically I saw little or no speedup regardless of how many CPUs I threw at it, which suggests that the bottleneck might be the memory system. I tried a few things - prefetching, for example - without much success.
Simon, you can investigate influence of memory speed to GC speed by manipulating memory timings on standard GC. for example, you write random-memory-access program, measure its slowdown when you change memory timings, then measure slowdown of GC with the same memory timing change another variant - try to increase CPU speed without changing memory speed. if this don't allow to make GC faster (including parallel GC) - this means that process is memory-bound -- Best regards, Bulat mailto:Bulat.Ziganshin@gmail.com