I'm not sure if i'm the only one having problems with these servers after like a map change - then the whole server seems to hang and you get thrown out or something. It also uses a different server mod than the standard one and i guess that's the issue. I assume it's Molgrum's own mod and that's why it's not changed to KTX.

However, i ping good to the servers and wouldn't mind seeing another mod used on them. Does anyone agree with me or am i just being a pain in the ass?

Last time I tried KTX 1.33 it ate up too much CPU, I don't know if that's fixed now. Regarding the server crash, it's a bug in the ftesv MVD streaming code, I'm debugging it right now and hopefully it gets solved pretty soon.
It was a while ago since I did anything to those servers, but now i changed the start map to dm4 and the countdown length to 15.

was that running ktx on fteqwsv or mvdsv?

if fteqwsv (is PR2 support up to date?), then try mvdsv and compare cpu usage
if mvdsv, then, what platform?
if osx, then please run using built in OS profilers and provide cpu trace
if linux, then please link against some profiler (e.g. google performance tools) and provide cpu trace

From what I remember, choice of server didn't matter (significantly).

The servers are restarted without any QTV connected to them, a temporary workaround that should prevent the crashes.

Are you going to give any trace so we can have a read?
What "too much CPU" means to you (roughly)?

I tried it with just me connecting now, couldn't see anything special but it might be worse when more players connect. How is the QVM implementation performance-wise?

Never did some real stress test of QVM or SO/DLL...
Might as well do that if I can figure how to do so.

Oldman, I'd be interested with some help on the topic

I did some stress-testing with Johnny_cz, 8 players shooting/moving (just basic +left;+attack) and 7 spectators:
mvdsv + ktx qvm: 22 % CPU, 8 RAM
fteqwsv + non-ktx: 20 % CPU, 4.1 RAM

Hard to say if it's fteqwsv or non-ktx that makes the performance better though.

.QVM takes 1½-2½ times more cpu time than .so/.dll and the memory consumption is totally flawed. MVDSV with 16MB heap takes ~23MB memory and with 32MB heap (default!) it takes ~39MB memory when idling.

With 2GHz K8 processor one player takes around 1% of the cpu using mvdsv + .so.

You can't even start to compare .QVM with .dat because their cpu-usage is on so totally different levels.

Hey, good time to define a protocol for testing CPU and memory use of the app

But .so/.dll has a drawback from a security POV, or do you have anything to say there?

*YAWN*

Ok as usual then, case closed.

Yes.

QW servers have been running "insecure" for the last 10 years, and look how we are being flooded by evil cheaters, damn them.

Or if you are referring to being able to hack on the shell like with ktpro, one word: CHROOT.



How long have you been on the scene, like a year or so? It's pretty evident that you worry about matters that are rather irrelevant. On the scene with this size and knowlege, would you think players spot cheaters in a nanosecond? (like they have done in the past)

even with .qvm you should be running with chroot

Soooo, back on the performance track (I'm not talking about security here, no need to call in the reinforcements)... Anyone with knowledge on how to measure use of cpu/mem of an application accurately?

deurk: setup http://cacti.net/ and http://www.rrdtool.org/ to get some graphs for your servers

Already have that, but I want something that records for a few minutes what is going on and allow analyse

Would be nice to compare servers and such, but I'm too lazy to do it