Very well put. My feelings exactly. Jeremy
Quoting Michael Ressen <[EMAIL PROTECTED]>: > > Increased CPU loads are the biggie, Eric. Figure it this way: You > size your servers and plan your capacity based on 3.1.1.0x versions, > which have been the standard for quite a while. Now, migrating to > 3.1.1.1x reduces your overall capacity by not just the ~20-25% that you > see in increased load, but can cause even bigger probs for those who run > multiple hlds on a physical box. > > For example, my biggest server is a 2x2.8 Xeon machine, using FreeBSD > 5.1, with hyperthreading enabled, and the HZ compiled in at 1000. All > extra stuff is stripped out. Since hlds is only single-threaded, the > scheduler is running tasks across one cpu at a time, although the > scheduler is alternating between phyical cpus to balance the load. Now, > here's my hlds command line: "./hlds_i686 -noipx -game cstrike +ip > 64.108.152.130 +port 27018 +maxplayers 32 +servercfgfile b1server.cfg > +map de_dust2 -pingboost 3 +sys_ticrate 1000 -tos". I have VAC, metamod > 1.16, hlguard 1.50 using aimbot, cvar, and file checks, and amx 0.9.3 > with 21 plugins. Watching my 'top' output right now here's what I get > at 10 sec intervals on cs_assault: > > CPU % (1 proc) # players > ------------------------------- > 74.22 26 > 71.09 25 > 77.34 25 > 80.47 27 > 82.03 27 > 82.81 26 > 83.59 26 > > If this was cs_office, or de_torn, she'd be pegged at 99%, and there'd > be some much higher pings in there. My only recourse is to turn the > pingboost down, which is going to result in adding +10ms or so to all > players and it won't be as smooth. Now imagine running an identical > server on the same box. With 3.1.1.0x I can, but 3.1.1.1x I probably > cannot. So, that puts me in a predicament; I can change my setup and > lower the load a bit, but my patrons won't like it. I'll lose people > doing that. On the other hand, to keep 2 servers running, at the same > level of performance, I now have to go out and spend another $2200 on > new hardware, plus space fees, additional router port, etc. The costs > to me for just one server are more than I am going to undertake, so you > can hopefully realize some of these guys here with 100 servers, who > might find they no longer have the capacity to handle the load. > They're out revenue. > > We do like all the things that you folks have put into the game, and > that's what's helped make it so successful, but we get frustrated when > we look at some of the other game engines out there and compare. > > For example, that same box that hosts my 32 player CS server also hosts > a 20-player rental. It also hosts a 20 player BF1942 server. Right > now, that server is full, running coral_sea, and the load is 30.47% (1 > proc). The efficiency of that binary is way ahead of the hlds binary, > and it uses 2.4 times the bandwidth player for player. > > I'm no super-duper programmer, I'm just some dumbass network engineer, > so I'm sure I'm gonna get flamed for this, but I'm also sure that there > will be more than a few people who will echo my statements here. > Anything that can be done to aid in the load issue we're facing is going > to be the most welcome relief to us. > > Thanks! > > Michael Ressen, > Michigan Burbs Network Administrator > > www.michiganburbs.com > _______________________________________________ > 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
