We've been running a lot of L4D servers for a while now, and with L4D2 we're again running a lot of servers as well (512 forks as of right now). We were originally running L4D primarily on Dual E5450 (3Ghz) based servers with 16Gb of RAM each in them, and we found that they would handle 64 forks with each core running at around 90-95% on average, and we found running them that close to maxed out for L4D didn't negatively affect in-game performance. Since then we've bought a few new servers based on the Nehalem chips, with the CPUs being the E5530s (Dual in each server) again with 16Gb of RAM (memory running at 1067Mhz).
So this brings us to L4D2, on the older E5450 hardware we've still got and use, we've found that L4D2 performance is virtually identical to that of L4D, in that we can again run 64 forks on those same servers with the each core sitting at 90-95%. Now with the Nehalems (E5530), we initially tried 64 forks and found with all servers full it was only sitting at around 50-60% CPU across all cores, so we cranked that up to 96 forks and all of the cores sit at around 80-85% CPU load on average (with all servers full), so realistically we could add some more forks if we wanted. Now the first thing people might suggest is that we're using a tweaked to hell kernel, but thats not the case at all. We're running them on a completely stock minimal install of CentOS 5.4, with nothing tweaked at all. We haven't tried using the 5530s for any other games as yet, but the performance increase that we've observed on L4D2 is quite impressive. Its also nice to see that Valve have been able to keep the Linux server CPU load about the same with L4D2 as it was with L4D. / Ben _______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
