Link Layer Adaptation for Ethernet

hi . i have ethernet connection with 36/36mbps download/upload . thats the speed i bought from isp but in dslreports i get like 90/90mbps somehow . should i use 90% of 36mbps or 90mbps ? also idk what to do in link layer adaptation on sqm . basically i need optimal settings for my internet . sry for bad english .

See this guide: https://openwrt.org/docs/guide-user/network/traffic-shaping/sqm-details

As stated there, use speeds from a speedtest with sqm off. Also, if your incoming connection is ethernet, select ethernet with overhead and enter 44. Test for bufferbloat. If score is good then youre done

1 Like

Tough question, if you get 90/90 to all targets independent of IP address you could/should configure SQM for 90 - safety-margin... you might want to talk to your ISP though, as they might have mis configured something (risking that they re-set your link's shaper to 36/36).

Regarding overhead, I concur with @randyrodriguez , if after reading that page you still have questions, please post them here.

1 Like

I only get 90/90 in speedtest program or sites ( like ookla , dslreports ) , if i download/upload file or something , the speed is 36/36 .

Most online speedtest now do multi connection speed tests. This means it will do multiple concurrent session for one speedtest. I have multiwan and my speedtests almost always use both connections. Many file transfers still use a single session so that may not saturate your connection. My isp likes to do free speed boosts for a month or two. I guess its to let you experience the higher speeds and tempt you to upgrade after they take it away. Maybe thats whats happening. Monitor your speed periodically and update your traffic shaping settings. At home, my connections are almost never saturated so i have sqm enabled but without traffic shaping. My bufferbloat tests usually give me an A or B+

Mmmh, in ookla's speedtest you can also select single stream tests, if you do, what rates do you reach?
But if your ISP has an exemption for speedtest sites, you probably should configure SQM for 36/36 with an appropriate safety-marging

https://www.speedtest.net/my-result/a/7799574641 - single stream test .
https://www.speedtest.net/my-result/a/7799579978 - multi stream test .

Odd, what does your ISP advertise as speed for your plan?


Text basically says up to 36mbps .

I guess I would then set SQM for 36*0.9 = 32.4 Mbps, or 32400 (since SQM expects rates to be in Kbps). According to baktelecom this is GPON, so I would guess that you need to account for 23-42 bytes of per-packet-overhead. And since I have no reliable way to heuristically measure this, please set the overhead to 42 Bytes which should be safe...

My connection coming with ethernet cat5e cable tho ( idk what kind of cable it is outside ) , but values should be pretty close i guess ?

This would be good if you want to set it and forget it, and if you’re happy with that speed. However, if you want to maximize what’s available, base your settings on the speedtests (as specified in the guide). I also suggest periodically turning off sqm and testing the speed in case it changes. I also recommend testing and tweaking to get the best results. Incrementally increase and decrease your settings and look at ping jitter to find the sweet spot. Aside from dslreports bufferbloat test, try doing it manually. Do a continuous ping then try to saturate the connection like with a speed test. Stop the ping and copy the results. Google ping jitter calculator. The first site has a tool where you can paste your ping results to get the jitter. The lower the value the better.

1 Like

All good points. BTW if you want jitter values, you can configure mtr to give jitter:

computer:~ user$ sudo mtr -ezbw -o LSBAWVGMXI -c 100 www.google.com
Password:
Start: 2021-10-11T22:11:07+0200
HOST: 123-1234567.local                                                                                       Loss%   Snt  Best   Avg  Wrst StDev Gmean Javg Jmax Jint
  1. AS6805   dynamic-2a01-0c23-91ba-4600-0000-0000-0000-0001.c23.pool.telefonica.de (2a01:c23                 0.0%   100   0.5   0.8   1.7   0.1   0.8  0.1  1.0  2.1
  2. AS6805   2a02:3001::11e                                                                                   0.0%   100  10.3  17.8 126.9  16.2  14.7 10.4 115. 163.5
  3. AS6805   2a02:3001::1b7                                                                                  39.0%   100   9.7  10.3  11.8   0.4  10.3  0.3  2.1  3.9
  4. AS6805   2a02:3040:0:10::1d                                                                               0.0%   100   9.8  10.7  22.1   1.2  10.6  0.6 11.8 20.6
  5. AS???    ???                                                                                             100.0   100   0.0   0.0   0.0   0.0   0.0  0.0  0.0  0.0
  6. AS???    ???                                                                                             100.0   100   0.0   0.0   0.0   0.0   0.0  0.0  0.0  0.0
  7. AS15169  2001:4860:0:1::21a2                                                                              0.0%   100   9.4  10.1  11.0   0.3  10.1  0.3  1.0  4.8
  8. AS15169  2001:4860:0:12e3::2                                                                             28.0%   100  10.4  11.0  15.3   0.6  11.0  0.5  4.8  5.4
  9. AS15169  2001:4860::1c:4001:ec9                                                                          14.0%   100  10.7  14.7 285.9  29.6  11.9  6.8 274.  8.8
 10. AS15169  2001:4860:0:1::1c5d                                                                             74.0%   100  10.1  10.9  11.6   0.3  10.9  0.4  1.1  4.7
 11. AS15169  ham11s01-in-x04.1e100.net (2a00:1450:4005:800::2004)                                             0.0%   100   9.4  10.0  10.8   0.3   9.9  0.3  0.9  4.1

but IMHO often just looking at the raw ICMP RTTs is enough to see whether a link suffers from bufferbloat or not :wink:

1 Like

Nice! Didn’t know that. Much easier. Thanks!