On Wed, Nov 26, 2008 at 11:09 PM, Milton Ngan <[EMAIL PROTECTED]>wrote:
> The CPU numbers are for a single dedicated server. There are definite CPU > spikes when the game initially starts and when large mobs get spawned. If > you watch the CPU usage over the course of an entire match you will get a > better idea of the spikes. I did this profiling using "top" in batch mode > sampling every second. It still isn't perfect, but it catches the spikes a > lot better than a 5 minute instantaneous sample or a 1 minute average. > > Versus doesn't change the CPU usage very much on the server side. The > physics simulation workload is the same, the AI is less and the network is a > little more. It does change network bandwidth (about double) due to more > clients. > > M. > > -----Original Message----- > From: [EMAIL PROTECTED] [mailto: > [EMAIL PROTECTED] On Behalf Of ics > Sent: Wednesday, November 26, 2008 1:18 PM > To: Half-Life dedicated Linux server mailing list > Subject: Re: [hlds_linux] hlds_linux Digest, Vol 9, Issue 199 > > This is single server instance? As a comparison (on full 4/4 server), we > saw around 15% average usage for Q6600 2,4GHz machine and at peak around > 24%, Q9550 2,83GHz only around 10%, peaked at 21% and dual core AMD > 4600+ 2,4GHz it is around 15%, peaked at 26%. I did not monitor for > long, only about 5 mins each time to time following cpu usage during the > tests i ran with them. Versus increases network traffic rapidly compared > to normal 4 player campaign so i would say that in versus cpu usage > would also be bigger. > > -ics > > Milton Ngan kirjoitti: > > While load average is a useful guideline, it does not necessarily reflect > CPU usage in all cases (e.g lots of I/O bound processes waiting to run). > > > > Also it is an average over time (60 seconds). It can't tell you if within > the last 60 seconds that your CPU's were over saturated for a second while a > number of the games spiked due to simultaneous mob spawns because that spike > will get averaged out. The only way to see that is to sample more > frequently, which unfortunately puts more load on the system. Or as someone > else suggested, try playing on the servers to get an idea of the servers > responsiveness. > > > > The CPU usage of a L4D dedicated server will range from 15% of a 2.5GHz > core to 60%, and average around 22% based on my observations. The more > servers you run, the more likely you are to run into simultaneous spikes and > saturate your CPU. If I can gather enough data on the events that cause > spikes I am sure I can work a statistical model for calculating the amount > of CPU headroom you need. > > > > M. > > > > -----Original Message----- > > From: [EMAIL PROTECTED] [mailto: > [EMAIL PROTECTED] On Behalf Of Ronny Schedel > > Sent: Wednesday, November 26, 2008 10:47 AM > > To: Half-Life dedicated Linux server mailing list > > Subject: Re: [hlds_linux] hlds_linux Digest, Vol 9, Issue 199 > > > > > > It depends on how many cores/CPUs you have. In the SMP world, you can run > 4 > > processes at 4 cores at the same time without any problem. Your load > average > > is 4 or less, depending on how much load they consume. > > > > Optimum: > > load average <= cores * CPUs > > > > Overload: > > load average > cores * CPUs > > > > So, optimum for one Quad Core CPU is: > > load average <= 4 * 1 > > > > This means not you can only run 4 L4D servers on one Quad Core, you have > to > > look how the load is, because the servers does not run all the time, they > > take some milliseconds sleep and other processes can run. Maybe you can > run > > 6 or 8. > > > > > > > >> Keep in mind that load average is an expression of queue depth. Some > >> types > >> of servers can handle deep queues with no visible impact shown to the > end > >> user. Gameservers are not that kind of animal. They require constant, > >> nearly instantaneous updates in both directions in order to have truly > >> smooth, seamless play. In the past, any time I saw load average above > one > >> I > >> saw an impact of some sort. > >> > >> A load average above 3 seems to manifest itself in L4D servers in some > >> unpredictable ways that can generally be ignored by most players. > >> > >> If you don't want to keep sending jarring notes that throw your players > >> off, > >> you really want to see your load average below 2. This should be true > >> regardless of the number of CPUs that you have in your box. Thus, my > >> comment in an earlier email about converting one of my public servers > back > >> over to Steam group only. Having 2 dedicated to just the Steam group > will > >> still let people fill up the public servers first and use the two > >> semi-private ones as relief. > >> > >> > >> [snip] > >> > >> > >> > >>> ------------------------------ > >>> > >>> Message: 3 > >>> Date: Wed, 26 Nov 2008 17:28:11 +0100 > >>> From: "Ronny Schedel" <[EMAIL PROTECTED]> > >>> Subject: Re: [hlds_linux] -fork configuration tip > >>> To: "Half-Life dedicated Linux server mailing list" > >>> <[email protected]> > >>> Message-ID: <[EMAIL PROTECTED]> > >>> Content-Type: text/plain; format=flowed; charset="iso-8859-1"; > >>> reply-type=original > >>> > >>> > >>> You have only 4 cores, with a load of 8 your server is overloaded. CPU > >>> usage > >>> does not matter these days, any new multicore CPU can handle Source > >>> servers. > >>> If you want lag free and smooth servers, your load average should not > >>> exceed > >>> your amount of cores. > >>> > >>> > >>> > >>>> Our load average goes up to about 8. Given that we're running 12 > >>>> instances > >>>> on a quad-core, this doesn't bother me, and certainly performance is > >>>> fine... > >>>> I've seen the blood spatter effect too, but I don't think that's > >>>> specifically performance related, it's just a glitch ;) > >>>> > >>>> > >> _______________________________________________ > >> 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 > > > > > _______________________________________________ > 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 > atop is better. Tyrael _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
