[bf1942] windows vs linux

[Tristan Gulyas]

Howdy,

>  From what you've said, my first guess would be network. What network
> cards are you using? If you use el-cheapo realtek cards, expect problems
> like this (although the fact that it works under Windows seems strange)
> My personal preference is the Intel Etherexpress cards, which are well
> supported under Linux and have always given me awesome performance. Some
> people swear by 3Com cards. My experience with them is that they're
> expensive and don't play well with Cisco routers (whose fault this is is
> left as an exercise to the reader). At any rate, YMMV.

We're using the onboard nForce2 MCP networking controller.  One of the
machines has a mobo with an onboard Intel gigabit NIC, so we use that on
that machine.

All machines exhibit the same issues.

It seems that using multiple processors (having one CPU handle the map
change spike) and SCSI disks sounds like the key but for a non-profit LAN
organisation, the cost of dual processor systems is prohibitive (our choices
these days are either Xeon, Athlon MP or Athlon 64).

> If it's not the network card, I'd start looking at kernel network
> configuration tweaking. There's a lot of settings you can tweak for
> optimal performance, and if you're not already compiling a kernel for
> the right architecture, that can help (the fact that you said it
> performs better under Windows suggests it could be a kernel optimization
> problem).

Perhaps I should have mentioned *CPU spikes* as well.  i.e. for some non-map
change periods, hlds_l shoots up to 70% of CPU..

No problems with Q3-based server code, and even bf1942 behaves itself most
of the time.

> You should also try asking on the hlds_linux mailing list, as there's a
> lot of folks on there who have a great deal of experience running the CS
> servers and can probably tell you what you should be expecting out of
> similar hardware.

I shall sub to that one shortly :)

.t