No subject


Wed Mar 11 18:56:50 EDT 2009


engine seems to use those extra ticks, and CPU usage seems to be
proportional to the tickrate. 1000Hz has often been seen to create 15%
cpu usage from empty BF servers (not good). We run all of our servers on
top priority (-20) to make sure that they aren't affected by any other
processes on the box, but we found that for BF we needed to set the
priority back to 0 else the load on the CPU is just too high under
1000Hz for other processes to get a decent share. BF1942 doesn't seem to
be affected by this either, which is good :-)

We haven't done any proper testing for this issue, so the above really
is through a vague bit of diagnostic fiddling, looking at results and
doing random things to try and fix them, and then a bit of speculation.
I'm sure that Steve will have a 50-page document with dozens of graphs
on the matter ;-)

---------------------------------------
Chris Adams
Fragzzhost

T (07005) 964 855
F (07005) 964 857
www.fragzzhost.com
 

-----Original Message-----
From: Neal Clayton [mailto:xayd at vae-victus.org] 
Sent: 31 October 2004 02:44
To: bf1942 at icculus.org
Subject: [bf1942] Punkbuster and performance, question/idea

Been alot of talk lately about BF and Punkbuster and the performance 
related issues that come with PB.

A thought that came to mind, which was an issue in older games and 
Linux/Unix servers, was the process polling timer setting in the 
kernel.  The default from forever(?) up through 2.4.xx Linux kernels was

100hz.  Back in the Unreal Tournament days we used to recompile kernels 
with this setting changed to 1000hz instead, which reduced kernel 
latency by a noticable amount (10-15ms), allowing for higher "tickrate" 
settings on the game server and noticably improved responsiveness for 
each player regardless of ping.

How does the BF server work with the system timer?  Would this setting 
help for BF as well?  Or does the BF server/engine have a hard-coded 
"tickrate" that wouldn't be affected by this?

http://kerneltrap.org/node/view/464/1528 for discussion/explanation
http://www.plumlocosoft.com/kernel/ 2.4.xx patches

Anyone tried this?

What about process priority?  Anyone tried manipulating it to see if it 
helps with overall performance?







More information about the bf1942 mailing list