Reply To: ASUS New DSL-AX82U Modem Router Combo – Wifi 6

[yakumo]

> manually allocating a set bandwidth per device is good but is time consuming .

I was testing altering bandwidth percentages by QoS priority rather than per device, but I think a slightly higher reduction on the overall bandwidth setting is a better solution, leaving the priorities at default. A FireTV Echo is causing real problems at times when in use with zero to 36Mbps in regular 3 second pulses even if set to a 20mbps device limit, it is more WAN disruptive than any P2P so far, if I can’t find a solution by router or by it’s own settings it may need replacing with a Shield.

Diabotical (free) and Quake Live are UDP based and excellent tools for testing latency. They have network graphs showing real time packet flow and packet loss and they’re very sensitive, sampling at a higher rate than ping/trace tool defaults, higher than many allow. As the latency is measured via the game’s usual UDP datagram, there is no risk of it being artificially shaped anywhere on route separately to the game traffic.

With Diabotical certain network conditions can also cause the game to apply automatic network restrictions, helping to prevent warp or abuse spoiling gameplay for other players, when these restrictions are in play the graph background turns red. The exact conditions that trigger this have not been specified but this is a third quality metric which can show even if latency and packet loss otherwise appear to be unaffected.

Testing involved playing while other consistent sources of traffic were seen or intentionally generated on the network, including SSL, OpenVPN, Netflix and Amazon streaming, a bittorrent of the latest iso from https://ubuntu.com/download/alternative-downloads .

Testing on an 18ms game server, the Adaptive QoS gave consistently higher latency in all heavy load test cases typically by 50ms or more and lost packets, the automatic network restriction was almost constant, with Traditional it was possible to eliminate that almost completely rarely gaining more than 7ms under heavy load (Ubunutu P2P) and without triggering the network auto restrictions, even if it they flared up for a moment on a large network spike it cleared very rapidly.

I need to investigate bufferbloat testing more, but dslreports speed test has a bufferbloat stat that was +200ms or more when using Adaptive QoS, if the experience in game had been better while the test was running I would have assumed it was entirely due to game traffic taking too much of a priority over the web, but the game was unplayable and the bufferbloat report also remained as poor when no game traffic was present so the QoS rule should have remained untriggered.

With Traditional rules and simply setting known game UDP ports to highest priority and the dl and ul main configuration set to 92% for sufficient reserve there is no need to alter the default bandwidth priorities and game experience is nearly flawless except under the most extreme loads, Adapative is simply not suitable for realtime gaming at all.

The Bufferbloat stat with this set up remained 2-5ms almost entirely after an initial spike at the start of the DSLreports traffic calmed.

TCP based games would not highlight problems as well due to TCP’s inherent ordering, retransmission, error recovery, only latency would remain an issue, but it is not often used for fast paced realtime games.

Thank you for the note on  fq_codel, I was wondering if something like that may be the case, now that is known I don’t need to experiment with custom firmware just yet.

Share the knowledge

You need to login in order to vote