There is probably no max value. This will render useless when ever you buy faster hardware. If there is they are just as ignorant as Bill Gates when he stated that "64K RAM is plently of memory, who would ever need more?"
Anyhow... Just bumping up HZ does not need to get a better system. HZ is how many switches a second the kernel should check which process to give its cpupower to. And should set accordingly to what your system should do. (Easily explained) Some system will work better with a low HZ, like 100, others much higher. I believe a critical number is 10000. Above that it is considered to do to much switching, so your system will never get faster and eventually slower if you set it too high due to that the kernel is switching too often and the system will not have enough power to run the applications. There are no good way to determine what you should set it to either. Since each OS is handling it differently. Most say that 1000 is a good average value though. But if we look into the future and we have 10GHz processors, HZ of 10000 might be considered to be the average HZ. Yet even more complicated are those OSes that have several HZ depending on whats being runned. Like having one for the kernel and userland having another. I believe some linuxes use 1500 for kernel and 1000 total for userland, though I have not checked it myself. And if I recall DragonFly BSD uses 1000 for kernel and letting userland have up to 100 per user (or was it process?). FreeBSD have just released a new version of their scheduler, which apparently is idling more than before and gives an over all better performance. What Im trying to say is: Hammering too much on the CPU might not give you a boosted system at all. The only way to find the perfect settings are simply changing values and see if it get better or worse. HZ is not the only option you can change. There are tons of other options you can change to get a smoother system, depending on what you are running. There are probably several thresholds that you can bump up or down. /Bjorn On Fri, 27 Jul 2007, Dan E wrote: > Just because you're telling the kernel that you want 10kHz doesn't mean that > the hardware can provide it. I'm pretty sure there is a max timer > resolution that the CPU can provide, and that is the limiting factor. Also, > don't do 125 tick; 100 is the maximum suggested by valve (issues arise from > >100tick. I'm sure someone else can expand on what I've said. > > -----Original Message----- > From: [EMAIL PROTECTED] > [mailto:[EMAIL PROTECTED] On Behalf Of Arne Guski > Sent: Friday, July 27, 2007 4:19 AM > To: [email protected] > Subject: [hlds_linux] Wondering where Ticks get lost (the old server fps > thing) > > Hi everyone, > i have been testing around with 1000fps servers for some time now and i > cant seem to get source running stable fps if i have more then one > server per cpu. > First of my test system is a C2D E6400 2gb ram running debian etch. > Currently running 2.6.22.1 with 4000hz (later more about that). > > It can handle 2 14+ Slot Servers fine, if i set the affinity to one core > per server i even get very stable 950fps (930avg over some days) with > setting tickrate to 125 and fps_max to 2000. > However if i dont taskset them to a specific core i still get "good" fps > but much more drops, which did not surprise me because ive red alot > about srcds having problems with multicores. > > Now what bothers me is that i cant get a third or even more servers > running stable high fps even though cpu on each core never exeeds 50%. > When i bind a second server on a core it will start up with much less > fps like the first one on that core (around 300 very jumpy). > If i start that third server with afinnity 3 (wich is random) it will > have better fps then when bound to a core (reaching 950 but falling down > to 300 very often). > > I also noticed that how well the server can hold its fps is map related, > these two servers i have running on the system are one dust2 and another > office only server. > Ive set up mrtg to monitor fps on both of these and the dust2 only > server has some glitches where fps drop while the office server almost > never had any fps glitches. > Now with all that fps dropping stuff making very little sense to me, it > could also be related to what server was started first or whatnot. > > Now the reason why i run 4000hz is that i thought the server might just > run out of ticks with a 2k kernel, wich does not apply though because it > is exactly the same with a 10k, 4k or 2khz kernel. > > So why is srcds acting like that and how can i get around that "one > server per core limit"? Note here that hlds can handle multiple 1000fps > servers fine with that setup until it runs out of cpu power. > > Looking forward to any suggestions, comments, idears on this one ... i > ran out of those. > > -- > Arne Guski > > > > _______________________________________________ > To unsubscribe, edit your list preferences, or view the list archives, > please visit: > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > > > _______________________________________________ > To unsubscribe, edit your list preferences, or view the list archives, please > visit: > http://list.valvesoftware.com/mailman/listinfo/hlds_linux > _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
