Peter, I found that performance with VMware was a bit spotty. Have you tried using a PV domain on Xen instead? That configuration should have a lower overhead. I would also recommend trying a 64-bit kernel instead of a 32-bit PAE, and a newer OS (CentOS 6, or even better Ubuntu/Debian). Two 24-slot TF2 servers and 5 L4D servers also could be a bit much for 3 virtual cores, depending on your CPU type (are you using Nehalem+? Older CPUs are much slower for virtualization), and your load average makes sense for having that many processes running.
-John On 9/9/2011 12:22 PM, Brian Menges wrote:
I really think the fact remains that the guest dom he's running is a bit overloaded; and running non-optimized settings. Namely the I/O is probably killing you (with whatever writes it may be achieving); Either run critical write portions in a RAM disk, optimize your guest, or run in full hardware. You have other options, but ultimately you're doing too much with one guest dom given the load you've displayed. ~B On Fri, Sep 9, 2011 at 12:19, John<lists.va...@nuclearfallout.net> wrote:We actually see very good performance here with optimized (using PV or HVM w/stubdoms and GPLPV, with various kernel and configuration tweaks) Xen setups, to within a small percentage of bare-metal. What hypervisor technology and speeds of machines have you been testing with? Are you allocating dedicated HT cores to each guest? -John On 9/9/2011 12:11 PM, Darren M wrote:I've noticed on cPGS users' machines that VMs / VPS of all flavors really struggle when it comes to running game servers well, even with a VM given 100% of all the resources of the dedicated box (for testing.) I strongly suggest to our users to save the VPS/VMs for mail or NS and let the game boxes be dedicated. Nothing taxes hardware in quite the same way that is also as prefectly monitored by so many people at the same time as game servers ^_^ ~darren On Fri, 2011-09-09 at 14:41 -0400, Andrew Caron wrote:Dont use a VM and with centos you need to build a custom kernel to have the fps stable on the servers. You have 3 cores, but your average load is 8, that means there is always something in queue to the CPU. Which also causes lag, your going to need to reduce the amount of servers, or get a better dedicated box. On Fri, Sep 9, 2011 at 2:37 PM, Peter Reinhold<peter_va...@reinhold.dk>wrote:Hi all, This question is not directly related to srcds, but may be more of a Linux question, but since there are a lot of experienced gameserver sysadms here, it seemed the logical place to ask. :) First off, my setup, so that is in place Running CentOS 5.6 in a ESXi VMWare, PAE kernel, three cores, 4GB RAM (I'm running a 1:1 on actual cores/virtual cores, so i'm not overextending the CPU) Two 24 man TF2 servers, and 5 forked L4D servers on this VM, nothing else. On to the problem: I've been hearing complaints from my players, and have experienced myself, small lag-stutters from time to time, where the entire server seems to stutter for at second or two, and then just continue. Pings stay normal, people aren't kicked or any such problems. So, i've started debugging (mostly monitoring with dstat and htop), and have found out something which I think is very strange, and may or may not be related to the issue at hand. TD2 servers have been more or less full the entire time (22-24 players at all times of monitoring), L4D have had 1-2 forks populated as well. Anyway, I am seeing load spikes, where load jumps from 1.5 to 3.5 in a single second tick in dstat, but the strange thing, there are no other metric that jumps at the same time, that could explain the large jump in load (I'm monitoring CPU, Mem, diskIO and process count, and there are no odd jumps there, no) Furthermore, this evening i'm seeing loads in the 8-9 range, but CPU utilization (usr/idle/wait), memory use, disk IO is the same as it was this afternoon, and its more or less the same number of players online (Give or take 5-7 players), and servers are as playable as they always are. Log files from dstat can be supplied, if anyone deems necessary. Does anyone know if this is srcds related, or my system thats just wonky (maybe reporting wrong load values?), or, possible know some metric that contribute to load, that I haven't monitored yet? Thanks in advance :) /Peter ______________________________**_________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/**mailman/listinfo/hlds_linux<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
_______________________________________________ To unsubscribe, edit your list preferences, or view the list archives, please visit: http://list.valvesoftware.com/mailman/listinfo/hlds_linux
