[bf1942] Server Utilization (Andreas)

[Paul Richards]

ok this is interesting I am testing BF 1.6 rc2 on a dual AMD 2800 2Gb box i
got sitting around and it runs very differently.

Idle CPU is 1.56-2% (which is fantastic)

It could be because its a faster box but the startup seemed alot quicker and
only climbed to about 45% cpu.

Andreas, is this simply because FreeBSD handles things differently ?

Cheers

Paul


----- Original Message ----- 
From: "David Snodgrass" <dave at pmmci.com>
To: <bf1942 at icculus.org>
Sent: Monday, February 02, 2004 4:30 PM
Subject: RE: [bf1942] Server Utilization


> Yeah that makes sense. My freebsd roots just encouraged me to enable SMP
to
> make HTT works - I'll compile my single boxes without SMP and see how it
> goes.
>
> -----Original Message-----
> From: Andreas Fredriksson [mailto:andreas.fredriksson at dice.se]
> Sent: Monday, February 02, 2004 11:24 AM
> To: 'bf1942 at icculus.org'
> Subject: RE: [bf1942] Server Utilization
>
>
> If the chip isn't SMP capable you're just compiling in synchronization
> overhead. For instance
> the Linux kernel spinlocks are compiled away without SMP defined.
>
> // Andreas
>
> > -----Original Message-----
> > From: David Snodgrass [mailto:dave at pmmci.com]
> > Sent: Monday, February 02, 2004 5:04 PM
> > To: bf1942 at icculus.org
> > Subject: RE: [bf1942] Server Utilization
> >
> >
> > On that note, FreeBSD taught me to compile SMP into my kernel
> > to take advantage of HTT on p4 and better systems. I carried
> > this habit onto linux, so - Even my single chip servers have
> > SMP kernels. Is that pointless/redundant on linux?
> >
> > Is there some environments you'd like to see/experiment with Andreas?
> >
> >
> >
> >
> > -----Original Message-----
> > From: Andreas Fredriksson [mailto:andreas.fredriksson at dice.se]
> > Sent: Monday, February 02, 2004 10:21 AM
> > To: 'bf1942 at icculus.org'
> > Subject: RE: [bf1942] Server Utilization
> >
> >
> > I've been trying to fix this bug since the first 1.4 release
> > but always come up empty handed. Profiling tells me that the
> > time isn't spent inside the BF server itself but in syscalls.
> > It might have to do with FPU exceptions or something equally obscure.
> >
> > In all honesty I wouldn't know how to produce a quick fix for
> > this bug.
> >
> > // Andreas
> >
> > > -----Original Message-----
> > > From: Paul Richards [mailto:noxis at clara.net]
> > > Sent: Monday, February 02, 2004 4:03 PM
> > > To: bf1942 at icculus.org
> > > Subject: Re: [bf1942] Server Utilization
> > >
> > >
> > > Andreas,
> > >
> > > Any idea on this SMP problem? As I am pretty sure the
> > > previous version didnt have this, I am running Gentoo with a
> > > SMP kernel and i just started up 1.6 for the first time,
> > > empty server and its idling at 20 - 39%. Is it going to be a
> > > possiblity to try to quickly patch this ?
> > >
> > > This is almost un-usable, all my unix/linux machines are SMP
> > > boxes does the CPU usuage scale when players join? or does it
> > > sit the same? If thats true then its no biggy and almost a
> > > help or does this mean that a full 24 player is going to be
> > > using like 90% cpu? If so this is a big step back in the
> > > linux server port :/
> > >
> > > Cheers
> > >
> > > Paul
> > >
> > > ----- Original Message -----
> > > From: "Andreas Fredriksson" <andreas.fredriksson at dice.se>
> > > To: <bf1942 at icculus.org>
> > > Sent: Monday, February 02, 2004 1:38 PM
> > > Subject: RE: [bf1942] Server Utilization
> > >
> > >
> > > >
> > > > I can't say why some servers are showing this behaviour,
> > > but it seems
> > > > to depend on your kernel configuration. Last summer I
> > received help
> > > > from a community member and we tested many configurations
> > and found
> > > > the culprit to be SMP kernels causing a massive CPU overhead.
> > > >
> > > > The same machine running a non-SMP kernel would not have
> > > the high-idle
> > > > problem.
> > > >
> > > > The carrier problem should be fixed in 1.6 since we have
> > > reverted the
> > > > carrier collision meshes back to the 1.31 versions which
> > > didn't have
> > > > the problem.
> > > >
> > > > // Andreas
> > > >
> > > > > -----Original Message-----
> > > > > From: Roland Kool [mailto:roland at cyso.nl]
> > > > > Sent: Monday, February 02, 2004 1:39 PM
> > > > > To: bf1942 at icculus.org
> > > > > Subject: [bf1942] Server Utilization
> > > > >
> > > > >
> > > > > Hello,
> > > > >
> > > > > Got some questions about CPU utilization behaviour I
> > > noticed since I
> > > > > installed the 1.6 patch on Linux. I've checked the forums @
> > > > > lightcubed but didn't see any threads about it yet.
> > > > >
> > > > > All idle servers (servers with no players) use between 15
> > > and 20% of
> > > > > CPU. Before 1.6 is was between 0 and 5%. That's the CPU
> > > util you see
> > > > > when starting 'top'. Caused by PB perhaps?
> > > > >
> > > > > If you use screen -r to resume your servers screen, I
> > > don't get all
> > > > > server console output. I have to press the enter and it
> > > looks like
> > > > > the CPU util reported there is around 100% which drops
> > > quickly back
> > > > > to 2-5%.
> > > > >
> > > > > Does 1.6 finally fix the collision meshes causing high
> > > util on maps
> > > > > with boats/carriers?
> > > > >
> > > > > Best regards,
> > > > >
> > > > > Roland Kool
> > > > > Thegamezone.nl
> > > > >
> > > >
> > >
> >
>
>