W2hCYK said:
I'm having some fun with it personally, so far!
I have been pulling some "experiments" here, mainly for the "dualcore/smp crowd"!
Just to see what we can do to operate MORE EFFICIENTLY/FASTER, using CPU affinity & process priority "tricks" & using switches for the consolemode/tty/charactermode clientware for this project (combining software characteristics & OS process/cpu affinity control)!
That's done to keep the cpu cycles dedicated to this process as high as possible (w/out undue "lockup", or the appearance of it happening)!
I.E.-> To keep the L2 cache clean via CPU affinity use (keeping SAME code running on a particular CPU, so its L2 cache keeps "sane" & doesn't change up code running on it for this process, via diff. threads taking over with their data).
L1-L2 cache amounts matter to distributed programming projects from what I have seen/heard over time (even from Steevo here in another thread, "FAH" is JUST LIKE SETI@Home in this regard)...
Keep the CPU L1-L2 cache "unpolluted" & theory says it will operate faster & this MAXES out whatever L1-L2 cache you have (regardless of amount).
My experiments above, so far @ least? Tend to "prove that/second that motion" so far... I gained a clean 20% over doing it w/out using CPU affinity &/or process priority settings!
W2hCYK said:
Fly3r is storming ahead. May I ask, what core are you working on?
Heh, gimme time - I'll catch him!
(I haven't brought my 2nd rig into this yet... that's going to be what does it for me, catching the leader!)
APK
P.S.=> I wish someone from the "SERIOUS ENTHUSIASTS" bunch @ the "FAH" forums would do a recompile, for the BEST optimization possible from this program for SPECIFIC CPU families & their instruction sets (e.g.-> AMD 3dNow/SSE1/SSE2/SSE3) &/or (e.g.-> INTEL MMX/SSE1/SSE2/SSE3) because I saw that happen w/ SETI@Home, & it made HUGE differences... 2x as fast many times, or nearly so! apk