[bf1942] preempt kernel and BF1942 server performance

[Mike Esler]

Disclaimer:  I only know what I have read here and there, and even then
I don't recommend you take what I say as fact :)
.  
I *think* that lowering response time doesn't necessarily mean that
overall system performance will increase or decrease, so I believe that
fps is not a valid test for latency. 

What's important is the BF1942 server not having to wait too long to get
CPU time to do what has to be done. For example, if the kernel is doing
something which takes 25ms to complete, then the BF1942 server will will
have to wait that long until it can do what it needs to do. This may add
some milliseconds to all players ping times.

I suppose a good benchmark might be to send a status packet to the
server, and see how long it takes for the server to respond.  This can
be done over a given period, and you can measure the average response
time, and the amount of variance in latency. I'd rather play on a server
that consistently gave me 50ms ping instead of one which fluctuates
between 25 and 75, and I think that is what the preempt kernel can do
for us, especially if there are other cpu/disk intensive processes
running alongside the server.

Of course, I may have this all wrong and the preempt patch may not do a
whole lot for us.  Perhaps the only people who will see a different in
latency are those who are running heavily loaded servers.  I'm looking
forward to some discussion on it from people who know more about the
subject than myself, and look forward to actually running a server with
and without it to observe the difference. 


On Fri, 2003-03-14 at 14:32, roger wrote:
> This should be noticeable by measuring the frame rates correct??
> 
> On Fri, 2003-03-14 at 14:00, Mike wrote:
> 
> > If I'm totally off base on my idea, please let me know ;)  The only test I 
> > have done (unscientific) was playing Quake 3 after kicking off a make 
> > bzImage modules after a make clean in the kernel src dir.  Overall, the 
> > game's jerkiness was less apparent and more consistent with the preempt 
> > kernel, making it very playable.  The stock 2.4.20 kernel 
> > did quite well, but the jerkiness was more apparent.
> > 
> > -- 
> > 
> > Mike Esler
> > mesler at themelee.org
> -- 
> Roger
> http://www.eskimo.com/~roger/index.html