I've done the CPU Core pinning thing, didn't help anything in my case. I don't think it is swapping cores that is the problem, it is just the ability to get CPU time in a stable and consistent rate with so many games swapping in/out of the CPU. I think less games with higher slot counts is the only way to maximize a 8-way Xeon setup.
I have a theory that FB RAM also makes the situation worse by adding additional latency to the overall system performance. We see a DPC latency of 3000us vs 30-50us on a server board vs non server board. I have setup a new server just for running 500-1000FPS CS/CSS games. It is DDR3 non FB RAM, 1 CPU and overclocked to the max, basically an enthusiast board setup vs a server setup. So far in testing it seems to be able to run more high speed servers than a 2x Xeon server board. So we are looking to use that server to run more lower end game servers that are not cranked to the max on the tickrate. > not it's not crazy. Running more than one cpu means the OS and programs > have to keep track of which core they are on, maintain cache coherency, > maintain the memory mappings..etc etc. If(when) the thread jumps to > another cpu then the whole dataset has to get rarranged. I would start > hard assigning servers to their own core and see if that helps out. > > _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
