QoS also limits speed to Router?

[eduroca]

I have QoS enabled on my N66U. The only rule enabled is that my machine is capped at 8Mb/s up/down (total WAN connection is 10 Mb/s up/down).

Ever since enabling it I've noticed that when I'm maxing out the 8Mb/s, accessing the router itself will become almost impossible, including the router's services like the FTP/SMB drive.

Normal LAN to LAN connections are unaffected (I can connect to my Linux box and copy files at full speed).

Any ideas why this might be happening or if it's even possible to avoid it?

Edit: N66U is running 378.56_2

 

[yorgi]

I can tell you why its because the ASUS QOS sucks like a big fat lemon :O

 

[sfx2000]

I have QoS enabled on my N66U. The only rule enabled is that my machine is capped at 8Mb/s up/down (total WAN connection is 10 Mb/s up/down).

Ever since enabling it I've noticed that when I'm maxing out the 8Mb/s, accessing the router itself will become almost impossible, including the router's services like the FTP/SMB drive.

Normal LAN to LAN connections are unaffected (I can connect to my Linux box and copy files at full speed).

Many of the traffic shaping/bandwidth moderating methods these days fall short - great for managing the WAN side, but nothing like the old-school Ubicomm stuff from back in the days before Qualcomm bought them...

It's not just Asus that has issues here - I've seen this with other vendors...

 

[A.D.]

I have QoS enabled on my N66U. The only rule enabled is that my machine is capped at 8Mb/s up/down (total WAN connection is 10 Mb/s up/down).

Ever since enabling it I've noticed that when I'm maxing out the 8Mb/s, accessing the router itself will become almost impossible, including the router's services like the FTP/SMB drive.

Normal LAN to LAN connections are unaffected (I can connect to my Linux box and copy files at full speed).

There seem to be several aspects to ASUS router behaviour under load which aren't necessarily what one would expect.

E.g. my AC66 in access point mode lets media streams for WiFi clients starve just because a single WiFi client is closer and requesting as much bandwidth as possible (by means of an iperf test).

The LAN to LAN traffic is not subjet to routing. It's switched (not even bridged, at least not in standard configurations). Also, data packets for established connections are forwarded in a hardware accelerated way, unless you enable some feature which prevents that.

Another user here reported a ping time of up to 1 second under load. The answer was "ping packets have low priority". Well. I think the system shows such behaviour both in the presence of lots of hardware accelerated data forwarding and high CPU load, and I personally would expect it to behave a bit more balanced.

It's a consumer grade device. I think it's optimised for certain use cases and it works best while no load limit is reached. I can live with that for access point operation. I currently don't think I'm going to buy an ASUS as a router.

 

[Nullity]

There seem to be several aspects to ASUS router behaviour under load which aren't necessarily what one would expect.

E.g. my AC66 in access point mode lets media streams for WiFi clients starve just because a single WiFi client is closer and requesting as much bandwidth as possible (by means of an iperf test).

The LAN to LAN traffic is not subjet to routing. It's switched (not even bridged, at least not in standard configurations). Also, data packets for established connections are forwarded in a hardware accelerated way, unless you enable some feature which prevents that.

Another user here reported a ping time of up to 1 second under load. The answer was "ping packets have low priority". Well. I think the system shows such behaviour both in the presence of lots of hardware accelerated data forwarding and high CPU load, and I personally would expect it to behave a bit more balanced.

It's a consumer grade device. I think it's optimised for certain use cases and it works best while no load limit is reached. I can live with that for access point operation. I currently don't think I'm going to buy an ASUS as a router.

You could try playing with the "wl" command as it has some WiFi QoS/traffic-shaping capabilities.

I often saturate my RT-N66U's WiFi (~200Mbit with my 2x2 clients) and I measure minimal to no change in latency. Saturating the WAN is different topic, but I assume that was what the "1 second ping" was referring to. I'd probably choose tomato since it has better QoS (like CoDel, IIRC). I use pfSense (with codel) as my router.


By the way, you differentiated that the device switches data but doesn't bridge it? What do you mean? Technically, a switch is just a multi-port bridge, right? https://en.m.wikipedia.org/wiki/Bridging_(networking)#Multiport_bridging

 

[Nullity]

I have QoS enabled on my N66U. The only rule enabled is that my machine is capped at 8Mb/s up/down (total WAN connection is 10 Mb/s up/down).

Ever since enabling it I've noticed that when I'm maxing out the 8Mb/s, accessing the router itself will become almost impossible, including the router's services like the FTP/SMB drive.

Normal LAN to LAN connections are unaffected (I can connect to my Linux box and copy files at full speed).

Any ideas why this might be happening or if it's even possible to avoid it?

Edit: N66U is running 378.56_2

I would assume some service like DLNA, SMB, Download Manager, etc was going haywire.

Have you tried running "top" or "ps" via SSH to see which service is being greedy?


QoS itself, especially with just 8Mbit to deal with should cause very little load.

 

[A.D.]

You could try playing with the "wl" command as it has some WiFi QoS/traffic-shaping capabilities.

Thank you for the pointer!

I often saturate my RT-N66U's WiFi (~200Mbit with my 2x2 clients) and I measure minimal to no change in latency.

Personally, I didn't see latency problems (or maybe it's the root cause) but throughput fairness issues.

you differentiated that the device switches data but doesn't bridge it? What do you mean? Technically, a switch is just a multi-port bridge, right? https://en.m.wikipedia.org/wiki/Bridging_(networking)#Multiport_bridging

From a system (black box) perspective you're right, but I was referring to software internals. I haven't come across a way to differentiate between at least between the four LAN ports in the default configuration on brctl level, so I suspect the idea is that LAN 1 to 4 traffic is not even checked against ebtables, which should make it even faster than traffic that's forwarded between the interfaces of the software bridge. (Access point mode here; WiFis, WAN and LAN are grouped in a software bridge).

(Although, from my experiments, it seems that if hardware acceleration is enabled, ebtables are only consulted when a connection is established and maybe at certain intervals afterwards -- so in that case, packets after the first one may not be slower through the software bridge than at LAN 1 to 4.)