ET server resource usage

(mkaminski) #1

Hey All…

I am going to be running a dedicated server at a data center, and ET will be the main game to run. I would like to know how many simultaneous 15-20 player ET servers I’d be able to run. I know that ET requires about 30MHz per player and 32MB of ram per game, but how do dual processors and SCSI hard drives take into account with performance?

The specs are as follows:

Dual Intel Xeon 2.8GHz
1GB ECC RAM
73GB SCSI hard drive
Linux

100mbit uncapped connection/2000 GB bandwidth (thats GB, not MB!)

I’d appreciate any advice and I apologize in advance if this is inappropriate for this forum!

(DG) #2

Et wont use dual procs, tho should be able to have n servers running off each cpu.

32mb seems awfy low to me, i think thats even the lowest usable setting just for the hunkmegs allocation, which might be a bit suspect for 20 players and/or mods, custom maps etc.

scsi… uh, the gameserver might initially load half a second faster?

i scribbled a small peice on et server requirements, but its pretty basic and not from too much experience since nobody wants to/can give data so no empirical research at all.

(mkaminski) #3

SCSI can sustain better data rates than IDE, so I assume it would be better for this type of multi-tasking.

Assuming increase the memory allocation to say, 56, what would be an estimate of how many simultaneous servers could run?

(chr0nicles) #4

SCSI can sustain better data rates than IDE, so I assume it would be better for this type of multi-tasking.

You wouldn’t gain anything ‘extra’ for using scsi instead of ide/sata running ET, this used to be different but with the current hdd ide/sata technology’s example (NCQ) performance equals and some setups even outperforms SCSI.

I don’t asume anything, i can give you statistical information and benchmarks, or you could do a google on the subject, since there’s lots of benchmarks out there defeating the “scsi is uber” myth.

in other words

DG = right, especially with the 32mb hunkallocation part.

(mkaminski) #5

I never said I would get increased performance or that SCSI was ‘uber’…I was inquiring as to whether SCSI would handle more load of multiple ET servers than IDE.

but with the current hdd ide/sata technology’s example (NCQ) performance equals and some setups even outperforms SCSI.

You were also assuming that the servers I’ll be hosting ET on will be using current up to date IDE drives. From my googling, I’ve found from different sources that SCSI is lower on processor usage and is better for processing more efficient I/O queries per second. This can be especially useful when stats stored in mySQL come into play.

According to the guide, I’d theoretically be able to support 9 ET game servers based on a 56 meg hunk usage and dual 2.8GHz processors… Does this sound plausible?

Thanks,
Mike

(Jaquboss) #6

beware also r_smp ( i think that this used for multiproc support in D3 and it is also may work in ET ) but i am really not sure :nag:

(SCDS_reyalP) #7

note the r_ prefix. That means it is a renderer cvar. Quake3 had SMP support in the renderer, not the server engine/gamecode. I seem to remember that even that is broken in RTCW and ET, but I’m not certain. In any case, it won’t help a dedicated server.

The whole SCSI vs IDE thing is pretty pointless. The only time an ET server is going to be disk IO limited is on startup and level changes. Any decently modern hardware and OS should have no trouble running the game and loading a file at the same time.

Note that hunkmegs does not account for all the memory used by ET, and your OS will want to have a fair bit for its own purposes.