Help on DSCP marking for gaming + SQM

Installing and Using OpenWrt Network and Wireless Configuration
41

Yepp, as expected, it is not cake that screws things up, maybe the DOCSIS plant is unhappy to see DSCP tagged ICMP packets?

I note I get less ICMP echorequest variations from the other side of the atlantic:

computer:~ user$ mtr -ezbw -c 100 172.217.13.99
Start: 2020-11-26T23:11:41+0100
HOST: xyz-1234567.local                                                           Loss%   Snt   Last   Avg  Best  Wrst StDev
  1. AS???    192.168.42.1                                                         0.0%   100    0.5   0.5   0.3   0.8   0.1
  2. AS6805   loopback1.0004.acln.06.ham.de.net.telefonica.de (62.52.201.201)      0.0%   100   13.8  15.9  13.2  52.8   6.7
  3. AS6805   bundle-ether11.0001.dbrx.06.ham.de.net.telefonica.de (62.53.7.108)   0.0%   100   14.2  14.2  13.7  14.8   0.3
  4. AS6805   bundle-ether1.0002.prrx.06.ham.de.net.telefonica.de (62.53.2.129)    0.0%   100   14.4  14.4  13.6  15.5   0.3
  5. AS???    ???                                                                 100.0   100    0.0   0.0   0.0   0.0   0.0
  6. AS15169  108.170.236.179                                                      0.0%   100   14.2  14.0  13.5  15.3   0.3
  7. AS15169  108.170.253.35                                                       0.0%   100   14.0  15.5  13.5  58.5   5.6
  8. AS15169  216.239.54.168                                                       0.0%   100   15.1  14.8  14.2  15.9   0.3
       [MPLS: Lbl 24064 TC 3 S u TTL 1]
  9. AS15169  172.253.50.110                                                       0.0%   100   19.0  19.3  18.7  21.9   0.5
       [MPLS: Lbl 25233 TC 3 S u TTL 1]
 10. AS15169  108.170.232.73                                                       0.0%   100   22.8  22.2  21.6  23.0   0.3
       [MPLS: Lbl 25721 TC 3 S u TTL 1]
 11. AS15169  209.85.142.167                                                       0.0%   100   27.4  28.3  27.0  69.1   4.9
       [MPLS: Lbl 29841 TC 3 S u TTL 1]
 12. AS15169  142.250.233.160                                                      0.0%   100   94.9  94.9  94.3  97.7   0.4
       [MPLS: Lbl 26849 TC 3 S u TTL 1]
 13. AS15169  142.250.46.67                                                        0.0%   100  103.3 103.7 103.2 104.7   0.3
 14. AS15169  108.170.251.49                                                       0.0%   100  103.1 103.0 102.4 103.6   0.2
 15. AS15169  108.170.231.59                                                       0.0%   100  103.5 104.1 103.3 106.7   0.5
 16. AS15169  yul02s04-in-f3.1e100.net (172.217.13.99)                             0.0%   100  103.1 103.0 102.4 103.7   0.2

This is through cake's best effort tin on a 100/32 shaped link...

user@computer:~$ ping -4 -c 100 -Q 184 172.217.13.99
[...]
--- 172.217.13.99 ping statistics ---

100 packets transmitted, 100 received, 0% packet loss, time 99142ms

rtt min/avg/max/mdev = 102.318/102.767/103.688/0.227 ms

even when forcing cake's highest priority tin the range is just 102.3-103.688ms not what you are seeing...

Note -Q 184 means decimal value for 8 bit TOS field which is equivalent to decimal 46 for the 6 bit dscp bitfield, which typically is called EF. And mtr actually understands -Q as well:

computer:~ user$ mtr -ezbw -c 100 -Q 184 172.217.13.99
Start: 2020-11-26T23:32:21+0100
HOST: xyz-1234567.local                                                           Loss%   Snt   Last   Avg  Best  Wrst StDev
  1. AS???    192.168.42.1                                                         0.0%   100    0.6   0.5   0.4   0.7   0.1
  2. AS6805   loopback1.0004.acln.06.ham.de.net.telefonica.de (62.52.201.201)      0.0%   100   15.9  17.3  13.1  55.1   9.0
  3. AS6805   bundle-ether11.0001.dbrx.06.ham.de.net.telefonica.de (62.53.7.108)   0.0%   100   14.3  14.2  13.6  15.2   0.3
  4. AS6805   bundle-ether1.0002.prrx.06.ham.de.net.telefonica.de (62.53.2.129)    0.0%   100   14.0  14.3  13.6  15.2   0.3
  5. AS???    ???                                                                 100.0   100    0.0   0.0   0.0   0.0   0.0
  6. AS15169  108.170.236.179                                                      0.0%   100   13.9  14.0  13.4  15.0   0.3
  7. AS15169  108.170.253.35                                                       0.0%   100   14.0  15.0  13.3  49.8   4.2
  8. AS15169  216.239.54.168                                                       0.0%   100   14.5  14.8  14.0  16.0   0.3
       [MPLS: Lbl 24064 TC 3 S u TTL 1]
  9. AS15169  172.253.50.110                                                       0.0%   100   19.3  19.5  18.5  43.1   2.4
       [MPLS: Lbl 25233 TC 3 S u TTL 1]
 10. AS15169  108.170.232.73                                                       0.0%   100   22.3  22.4  21.6  30.7   0.9
       [MPLS: Lbl 25721 TC 3 S u TTL 1]
 11. AS15169  209.85.142.167                                                       0.0%   100   27.2  27.7  26.9  37.9   1.1
       [MPLS: Lbl 29841 TC 3 S u TTL 1]
 12. AS15169  142.250.233.160                                                      0.0%   100   95.0  95.4  94.4 114.8   2.7
       [MPLS: Lbl 26849 TC 3 S u TTL 1]
 13. AS15169  142.250.46.67                                                        0.0%   100  104.0 103.8 103.1 104.7   0.3
 14. AS15169  108.170.251.49                                                       0.0%   100  102.8 103.0 102.4 104.0   0.3
 15. AS15169  108.170.231.59                                                       0.0%   100  103.8 104.3 103.3 112.4   1.3
 16. AS15169  yul02s04-in-f3.1e100.net (172.217.13.99)                             0.0%   100  103.1 103.0 102.5 103.9   0.2

Note, to convert 6bit DSCP to 8 bit TOS values, one needs to "push" all bits two bits to the right, which is equivalent to multiplication by 4:
EF: dscp decimal: 46 -> 46 * 4 = 184 TOS decimal.

42

Yes this is very likely. With Cable/DOCSIS the entire neighborhood shares the same signal, so definitely you can get congestion in your local neighborhood.

You could TRY to drop your speed a lot more... like make it half of what you are supposedly provided... 45/12. you might see a lot more consistency. Honestly for gamers bandwidth doesn't matter above about 1MTU/ms ~ 12Mbps and consistency in ping is way more important.

1 Like
43

You could test that by adding the wash keyword to SQMs eqdisc opts, to re-map all packets DSCPs to 0 AFTER cake used the previous value to assign to a specific tin, the idea is to avoid leaking DSCPs upstream.

44

I might try that!
but at this point it does seem like an issue with something down the line, probably not 'ping' command's fault

45

Blockquote You could TRY to drop your speed a lot more... like make it half of what you are supposedly provided... 45/12. you might see a lot more consistency. Honestly for gamers bandwidth doesn't matter above about 1MTU/ms ~ 12Mbps and consistency in ping is way more important.

yea I've tried to limit my bandwidth really aggressively. But it's not really distinguishable by reading the pings and the lag in game is still there for sure. : (

think it's just something out of my hands now

1 Like
46

This topic was automatically closed 10 days after the last reply. New replies are no longer allowed.