MT GC speed

BlitzMax Forums/BlitzMax Programming/MT GC speed

JoshK(Posted 2009) [#1]
My first real attempt at multithreading produced results where the MT version was about 50% slower. Mutexes were minimal and the thread and core count was 4.

Has anyone else experienced results like this? Right now it appears like maybe the extra work the GC has to do nullifies whatever advantages you might get by using more CPU cores.


GfK(Posted 2009) [#2]
I've only done basic testing with the threading module so far but I haven't noticed any speed problems yet. I guess it depends on how many threads you have, and what those threads are trying to do. Although, Mark did say this Helena-Bonham-Carter garbage collector or whatever the hell its called, is slower than the BRL one.


Otus(Posted 2009) [#3]
Same as what I noticed. CPU limited test programs with threads were slower than without (on a dual core). Other tasks like networking or file IO might be faster on threads.


JoshK(Posted 2009) [#4]
I found I got better speeds when I used like corecount*8 threads. That was interesting.


ImaginaryHuman(Posted 2009) [#5]
Maybe the mutexes are putting the threads on hold for too long and the more fine-grained you get in terms of how many threads can be inserted into those timeslots the more use you can make of those stuck phases?