Re: [tsvwg] quick update on gaming traffic rates

[Sebastian Moeller <[e-mail]>]

Hi Greg,

see [SM] below:


> On Sep 25, 2023, at 19:50, Greg White <[e-mail]> wrote:
> 
> Hi Sebastian,
> 
> Thanks for sharing these data points.  
> 
> I think that both CSGO and CS2 are the type of application that would benefit from NQB classification in networks that support the PHB.  Both of these are very popular and very latency (and latency variation) sensitive games.  So, ideally the NQB sender requirements would be written such that these applications can take advantage of NQB marking. 
> 
> For each of these games, was there only a single microflow involved, or was the traffic split across multiple microflows?  Could you share the pcaps for further analysis?
> 

I asked for and got consent to share the packet captures, these are from CSGO and a (recent closed) beta version of CS2, both are captures of around 1 minute game play one taken a more quiet network and a second with an upload & download stressor software running on the same host as the game this was intended to simulate other users in the network). The router was configured using sqm-scripts/cake for both down- and up-link, DSL speed was around 20/1 Mbps and the (cake) shaper was set to 17000/950 Kbps (details revealed in the linked forum thread).

The following link should be usable for the next 7 days only
https://easyupload.io/4maewv

In later experiments the user noted that setting the upload shaper to >= 256 Kbps results in playable CSGO, while CS2 apparently became playable only at 512 Kbps:

Here is a direct quote:
"I used sqm and gave 128kb upload for csgo to see exactly how much this game requires.
128kb was lag and unplayable
256kb was good and playable
did the same to CS2
128/256kb both were so bad lag and packet loss
512 is where things start getting better but some spikes from time to time
which means this new game requires more than half mb to be really playable."

The packet capture has not been anonymized in any way so please use it carefully and responsibly, as I am sure you will even without me mentioning it.

Regards
	Sebastian

P.S.: The user noted that the CS2 release version from yesterday still behaves like the beta regarding lag so likely the current behaviour is still similar to the captures.
P.P.S.: If your analysis finds anything the user might change to improve the situation please let me know and I will pass this on, or you can directly address the user in the linked OpenWrt forum thread over at: https://forum.openwrt.org/t/csgo-vs-cs2-using-sqm/171272?



> -Greg
> 
> 
> 
> On 9/24/23, 1:57 PM, "tsvwg on behalf of Sebastian Moeller" <[e-mail] <mailto:[e-mail]> on behalf of [e-mail] <mailto:[e-mail]>> wrote:
> 
> 
> Dear list,
> 
> 
> in the course of discussing the NQB draft we at one point talked about what kind of traffic rates we expect to see for common first person shooter type online games.
> I have got hold of a few short packet captures of 1 minute game play in the popular? game counter strike global offensive (CSGO) and it's successor couterstrike2 (CS2). These captures are not mine, but were taken by an active player...
> 
> 
> Here are the rates I extracted from the packet captures:
> 
> 
> CSGO:
> upload rate: 97 Kbps (@~64Hz)
> download rate: 430 Kbps (@~64Hz)
> 
> 
> CS2:
> upload rate: 375 Kbps (@~64Hz)
> download rate: 1073 Kbps (@~105Hz)
> 
> 
> 
> 
> I offer this as data for discussing the NQB requirements... with the current recommendation of <= 500 Kbps, CS2 would not be eligible, especially if the upload rate is well below the 50 Mbps the draft mentions. I also note that in the download direction there are "ticks" with single packets, but also with 2 or 3 packets back to back which might also challenge the current smoothness requirements.
> 
> 
> In case someone wonders, this is from the OpenWrt forum thread https://forum.openwrt.org/t/csgo-vs-cs2-using-sqm/171272? <https://forum.openwrt.org/t/csgo-vs-cs2-using-sqm/171272?> and is based on problems playing CS2 on a 20/1 link once there is some competing traffic in the link, the link in question is using cake as traffic shaper, AQM and fq scheduler. (Here part of the problem that the required capacity share for smooth game play is above the fair share, but for the discussion of what do modern games actually require this data still seems relevant).
> 
> 
>