Merlin 384.17 Adaptive QoS on AC86U with Comcast Gigabit service

[maxbraketorque]

I found out that switching to the Comcast Gigabit service was cheaper than my current internet service plan so I upgraded this week using a Netgear CM1150V modem. When I hooked my laptop directly to the modem and ran the Fast.com speed tests, the computer easily hit the 1 gbps inbound and 35 mbps outbound, but buffer bloat was monsterously high.

With my AC86U hooked up running Adaptive QoS on fq_codel option with manually set 980/35 limits, speeds on Fast.com running from a wired laptop tended to top out at 650-750 mbps on the inbound and 33-35 mbps on the outbound with loaded latency values of ~50-70 ms. I was rather surprised to see speeds topping out a 750 mbps with the AC86U in the path. The sub-1 gbps download speed is not a big because I doubt I'll ever need 1 gbps, but it was still surprising because I figured that the AC86U would easily handle this. Dropping the limits to 650/31 in Adaptive QoS dropped the fully loaded inbound and outbound latency values to 20-30 ms, and the wired laptop could easily hit the 650 mbps limit download.

Anyhow, just wanted to provide a data point.

 

[K-2SO]

I would turn QoS off completely. Your bufferbloat is high only in tests. Near line saturation speeds. How often you are going to hit Gigabit? How often you upload large data? Don't fix things if not broken.

 

[maxbraketorque]

I do send a fair amount of data offsite because of work and bit torrent. I'm definitely not sitting at 35 mbps all day or even for part of the day, but I figure that bursts of outbound data can still temporarily induce high latency which I don't want during VoIP and video conferencing.

However, I decided to try your suggestion. Turning off QoS doesn't give higher speeds in the tests, and buffer bloat at saturation on the outbound is high with QoS disabled. For kicks, I also tried disabling Traffic Analyzer and then Parental Controls which are the only other features I have enabled besides OVPN servers and clients on the router that I won't ever disable. Speed test results still peaked at 750 mbps on the inbound, and there was substantial buffer bloat on the outbound.

So I can't say I see any value in disabling QoS since it can mitigate buffer bloat during high outbound traffic.

 

[K-2SO]

Test it for some time. Forger DSLReports for now. It may not tell the whole story. Experiment with QoS only if you have issues. You have Merlin, use DNSFilter for parental. Traffic Analyzer is buggy, IMHO.

 

[maxbraketorque]

There's no obvious speed hit for having Traffic Analyzer and Parental Controls enabled, and I need those to help regulate my daughter's internet usage.

 

[K-2SO]

Traffic Analyzer likes to quit working. I was just about to update Asus 81792 thread. Does this on Asus and Merlin on my routers. It's like this forever. You may have better luck.

 

[maxbraketorque]

Its been working fine for me. Just used it the other day too figure out that my daughter has been using my wife's phone to watch movies after going to bed. I just wish TA had more filtering options.

 

[K-2SO]

Experiment with Adaptive QoS and FreshJR. Many people say it's working good. I can't share max throughput numbers. My ISP is ~300Mbps. My bufferbloat control is Traditional QoS with sfq. Works for me A/A+ always. No QoS works for me too. No family complaints, no stutter on anything. I don't pay too much attention to QoS. My kids are 16+.

 

[ColinTaylor]

For kicks, I also tried disabling Traffic Analyzer and then Parental Controls which are the only other features I have enabled besides OVPN servers and clients on the router that I won't ever disable. Speed test results still peaked at 750 mbps on the inbound, and there was substantial buffer bloat on the outbound.

It's unclear to me whether you are performing your speedtest through the VPN client.

 

[K-2SO]

It's unclear to me whether you are performing your speedtest through the VPN client.

The router can do 200+ OVPN. He's getting 750. Must be WAN.

 

[maxbraketorque]

Experiment with Adaptive QoS and FreshJR. Many people say it's working good. I can't share max throughput numbers. My ISP is ~300Mbps. My bufferbloat control is Traditional QoS with sfq. Works for me A/A+ always. No QoS works for me too. No family complaints, no stutter on anything. I don't pay too much attention to QoS. My kids are 16+.

I am using FreshJR for improved control over bandwidth prioritization. The added control is nice.

It's unclear to me whether you are performing your speedtest through the VPN client.

Not through any VPN clients.

 

[K-2SO]

The added control is nice.

What about your bufferbloat fight?

 

[maxbraketorque]

What about your bufferbloat fight?

Sorry I guess I didn't explain well. I've been using FreshJR almost the entire time that I've been using QoS. Setting the manual inbound/outbound limits seems to have the greatest effect on the degree of buffer bloat.

 

[K-2SO]

Sorry I guess I didn't explain well.

You did. My multitasking is failing.

 

[maxbraketorque]

Well, I guess there's some learning curve happening. I decided to run test speeds again this morning when I could reboot the router between feature changes without annoying anyone. First I verified again that I could get 1000/35 with my laptop directly connected to the modem. Then I connected that same computer to my AC86U followed by disabling Parental Controls, Traffic Analyzer, and QoS, and then rebooting the router. Somewhat to my surprise, I got 1000/35 again. Next I enabled Parental Controls and Traffic Analyzer followed by another reboot. This time I got 980/35 during multiple tests which is essentially full speed. Next was to add QoS set to 1000/35. This time inbound speeds reached about 900 mbps at best and generally peaked at 800-850 mbps for multiple tests. So I found that QoS definitely slows the inbound, and I also found that router reboots are required to truly disable some features.

As far as outbound buffer bloat goes, it wasn't as bad this morning with QoS disabled. Latency would generally hover at 50-90 ms at outbound saturation but would occasionally spike to 100+ ms. In comparison, with QoS enabled, latency during max outbound would be 20-30 ms, and unloaded latency is ~10 ms. I'm not sure why latency with QoS disabled wasn't so bad this morning. I'll have to try again this afternoon when internet usage on the West Coast increases.

 

[K-2SO]

Happy life begins. One less trouble. Free DSLReports test unavailable.
"Due to bandwidth costs the previously free speed test is now restricted to registered users with some history on the site"

 

[ColinTaylor]

Happy life begins. One less trouble. Free DSLReports test unavailable.
"Due to bandwidth costs the previously free speed test is now restricted to registered users with some history on the site"

Ironic. It was only a couple of days ago that I was in two minds as to whether I should make a post mentioning that almost all of their servers outside of the US have been offline since the beginning of the year. And of the few that were online not all of them did buffer-bloat tests.

 

[K-2SO]

since the beginning of the year

Oh, I feel privileged now. It was working last week. 45km away from the US border.

 

[bluzfanmr1]

Well, I guess there's some learning curve happening. I decided to run test speeds again this morning when I could reboot the router between feature changes without annoying anyone. First I verified again that I could get 1000/35 with my laptop directly connected to the modem. Then I connected that same computer to my AC86U followed by disabling Parental Controls, Traffic Analyzer, and QoS, and then rebooting the router. Somewhat to my surprise, I got 1000/35 again. Next I enabled Parental Controls and Traffic Analyzer followed by another reboot. This time I got 980/35 during multiple tests which is essentially full speed. Next was to add QoS set to 1000/35. This time inbound speeds reached about 900 mbps at best and generally peaked at 800-850 mbps for multiple tests. So I found that QoS definitely slows the inbound, and I also found that router reboots are required to truly disable some features.

As far as outbound buffer bloat goes, it wasn't as bad this morning with QoS disabled. Latency would generally hover at 50-90 ms at outbound saturation but would occasionally spike to 100+ ms. In comparison, with QoS enabled, latency during max outbound would be 20-30 ms, and unloaded latency is ~10 ms. I'm not sure why latency with QoS disabled wasn't so bad this morning. I'll have to try again this afternoon when internet usage on the West Coast increases.

I recommend using FreshJR_QoS if you are going to use QoS. I've found that it works better as opposed to using QoS alone.

 

[K-2SO]

I recommend using FreshJR_QoS if you are going to use QoS.

I've been using FreshJR almost the entire time that I've been using QoS.