Newell about CS:S network traffic

L

Lancelot

Guest
http://www.hl2fallout.com/forums/index.php?showtopic=2030

I measured the bandwidth demands for GoldSrc & Source with 16 bots on de_dust. This should be realistic since bots generate the same network traffic as human players do. I used a network monitor to get the exact traffic data for a process. Unfortunatly the tool doesn't include the UDP header size, so I had to add 0.55 kB/sec to all results. The resulting bandwidth per client is:

GoldSrc : 2.72 kB/sec
Source : 2.85 kB/sec

This is pretty good since we added quite a few physics entities in de_dust. I made an extra testrun where I disabled physics in Source and got 2.48 kB/sec.

It's tried to measure spikes somehow and compared packet size distribution. Source engine packets are a little bit bigger, but large packets are rare (about 7% for packets > 256 bytes).


setup: 2 hosts, measured 10 minutes per configuration
network settings: rate 10000; cl_updaterate 20; cl_cmdrate 30
GoldSrc: CS:CZ, de_dust_cz
Source: CS:S, de_dust


total in bytes total out bytes avg in kB/Sec
-----------------------------------------------------------------------------
GoldSrc testrun 1 : 1,357,534 443,948 2.21
GoldSrc testrun 2 : 1,306,061 412,817 2.13

Source testrun 1 : 1,498,837 436,307 2.44
Source testrun 2 : 1,323,526 429,419 2.15

Source no physics : 1,184,831 430,957 1.93

(all values without UDP header)


Packet sizes distribution in bytes:

GoldSrc Source
--------------------------------------
< 64 1.6% 1.7%
64 - 127 61.2% 56.1%
128 - 255 32.7% 35.0%
256 - 511 4.3% 6.9%
511 - 1023 0.1% 0.2%
> 1024 0.1% 0.1%
Click to expand...
 
In other words, the netcode itself is smoother than the HL1 netcode, but the physics stuff brings it up level with the current CS netcode, which is good.
 
You must log in or register to reply here.
Back
Top