FYI/A: Bufferbloat 101

xiaofan · Mar 29, 2024

Two out of the three servers mentioned in https://flent.org/intro.html#quick-start are still up. But the speed seems to be slow from Singapore. I also tested Sydney Starlink Server.
https://blog.cerowrt.org/post/flent_fleet/

Bash:

mcuee@debian12vmn100new:~/build$ lsb_release -a
No LSB modules are available.
Distributor ID: Debian
Description:    Debian GNU/Linux 12 (bookworm)
Release:        12
Codename:       bookworm

$ flent rrul -p all_scaled -l 60 -H netperf-west.bufferbloat.net -t flent_rrul_west -o flent_rrul_west.png

$ flent rrul -p all_scaled -l 60 -H netperf-west.bufferbloat.net -t flent_rrul_west -o flent_rrul_west.png

$ flent rrul -p all_scaled -l 60 -H sydney.starlink.taht.net -t flent_rrul_sydney_starlink -o flent_rrul_sydney_starlink.png

hwzlite · Mar 30, 2024

xiaofan said:
@hwzlite

I guess by Mr Dave you mean Mr Dave Täht, right?

Yes, he is the one who co-founded the Bufferbloat Project , the birth of FQ-CoDel, Cake SQM

xiaofan · Mar 30, 2024

hwzlite said:
Yes, he is the one who co-founded the Bufferbloat Project , the birth of FQ-CoDel, Cake SQM

Thanks. He is busy with LibreQoS now.
https://github.com/dtaht

I am wondering if he can help to host some crusader servers.

https://github.com/Zoxc/crusader/issues/14

xiaofan · Mar 30, 2024

Some wired test results from Ubuntu 22.04 Virtual Machines running under PVE 8.0 (just using Linux bridge, no passthrough), Intel N100 Mini PC with quad 2.5G Intel I226-V NICs.

No SQM rules enabled if for the two OpenWRT virtual routers.

There is a mis-match between LAN (2.5G connection) versus WAN (1Gbps internet) in these test cases.

1) Behind Asus RT-AX86U router -- A
Singtel ONT -- TL-SG105E smart switch -- Asus RT-AX86U WAN
RT-AX86U 2.5G LAN -- Intel N100 PVE 8.0 Ubuntu 22.04 VM
https://www.waveform.com/tools/bufferbloat?test-id=b5a6a630-f442-4fc4-a5aa-a31021160dcc

2) Behind OpenWRT virtual router 1 -- A
Singtel ONT -- TL-SG105E smart switch -- OpenWRT virtual router 1 WAN (2.5G) on Intel N100 PVE 8.0
Virtual router 1 2.5G LAN -- Intel N100 PVE 8.0 Ubuntu 22.04 VM
https://www.waveform.com/tools/bufferbloat?test-id=f22fa835-4ef4-4511-af52-0ae5c77263b2

3) Double NAT, behind Asus RT-AX86U + OpenWRT virtual router 2 -- A
Singtel ONT -- TL-SG105E smart switch -- Asus RT-AX86U WAN
RT-AX86U 2.5G LAN -- OpenWRT virtual router 2 WAN (2.5G) on Intel N100 PVE 8.0
Virtual router 2 2.5G LAN -- Intel N100 PVE 8.0 Ubuntu 22.04 VM
https://www.waveform.com/tools/bufferbloat?test-id=b9cc36a2-a555-4160-8470-4069c605eda1

4) Using an Ubuntu 22.04 VM with all the above three links (three network adapters) -- A+ and A
https://www.waveform.com/tools/bufferbloat?test-id=ce6d70d1-7530-4cdc-ac04-e5cb307b40a1
https://www.waveform.com/tools/bufferbloat?test-id=54c7c1dd-64db-4253-961f-08371781c247

As mentioned by brother @hwzlite, Linux with more recent kernel version will have fq_codel support built-in.

Bash:

mcuee@ubuntu2204vmbr0:~/build/speedtest$ uname -a
Linux ubuntu2204vmbr0 6.5.0-26-generic #26~22.04.1-Ubuntu SMP PREEMPT_DYNAMIC Tue Mar 12 10:22:43 UTC 2 x86_64 x86_64 x86_64 GNU/Linux

mcuee@ubuntu2204vmbr0:~/build/speedtest$ tc -d qdisc
qdisc noqueue 0: dev lo root refcnt 2
qdisc fq_codel 0: dev ens18 root refcnt 2 limit 10240p flows 1024 quantum 1514 target 5ms interval 100ms memory_limit 32Mb ecn drop_batch 64

Same for OpenWRT.

Bash:

root@OpenWrt:~# tc -d qdisc
qdisc noqueue 0: dev lo root refcnt 2
qdisc fq_codel 0: dev eth0 root refcnt 2 limit 10240p flows 1024 quantum 1514 target 5ms interval 100ms memory_limit 32Mb ecn drop_batch 64
qdisc fq_codel 0: dev eth1 root refcnt 2 limit 10240p flows 1024 quantum 1514 target 5ms interval 100ms memory_limit 32Mb ecn drop_batch 64
qdisc noqueue 0: dev br-lan root refcnt 2
qdisc noqueue 0: dev wg_lan root refcnt 2

xiaofan · Mar 30, 2024

Here are some wired test results using Windows 11 Acer Laptop and Ugreen USB 3 to 2.5G adapter.

There is a mis-match between LAN (2.5G connection) versus WAN (1Gbps internet) in these test cases.

The two OpenWRT virtual routers are running inside Proxmox PVE 8.0 on Intel N100 Mini PC with quad 2.5G Intel I226-V NICs.

No SQM rules enabled if for the two OpenWRT virtual routers.

Still the results may not be real as the Speed is low for the first two tests, much lower than the 940+ Mbps OOkla SpeedTest results.

1) Behind Asus RT-AX86U router -- A+
Singtel ONT -- TL-SG105E smart switch -- Asus RT-AX86U WAN
RT-AX86U 2.5G LAN -- Acer laptop
https://www.waveform.com/tools/bufferbloat?test-id=f34d3f2e-c01d-47ee-b0cd-5f2d478e1546

2) Behind OpenWRT virtual router 1 -- A
Singtel ONT -- TL-SG105E smart switch -- OpenWRT virtual router 1 WAN (2.5G port)
OpenWRT virtual router 2.5G LAN -- Acer laptop
https://www.waveform.com/tools/bufferbloat?test-id=a1968217-f78a-456c-acae-217bd38ed00e

3) Double NAT, behind Asus RT-AX86U + OpenWRT virtual router 2 -- A+
Singtel ONT -- TL-SG105E smart switch -- Asus RT-AX86U WAN
RT-AX86U 2.5G LAN -- OpenWRT virtual router 2 WAN (2.5G port)
OpenWRT virtual router 2 2.5G LAN -- Acer laptop
https://www.waveform.com/tools/bufferbloat?test-id=9060beca-dd38-447d-9837-281b2e0f0fc5

xiaofan · Mar 30, 2024

In the previous two wired test cases, OpenWRT virtual router 1 gets A.

How to get A+? Limit the upload speed to 200Mbps is one potential solution judging from the test results.

Using CAKE in the following tests.

1) Acer Windows 11 laptop with Ugreen USB 3 to 2.5G adapter, behind OpenWRT virtual router 1 (A+)

First test to show that the SQM rule works fine.

Bash:

PS C:\work\speedtest\ookla-speedtest-1.2.0-win64> .\speedtest.exe -s 13623

   Speedtest by Ookla

      Server: Singtel - Singapore (id: 13623)
         ISP: Singtel Fibre
Idle Latency:     2.58 ms   (jitter: 0.31ms, low: 2.15ms, high: 2.87ms)
    Download:   920.84 Mbps (data used: 450.2 MB)
                  3.55 ms   (jitter: 0.64ms, low: 1.88ms, high: 5.04ms)
      Upload:   192.71 Mbps (data used: 161.5 MB)
                  1.96 ms   (jitter: 0.51ms, low: 1.64ms, high: 24.64ms)
 Packet Loss:     0.0%
 Result URL: https://www.speedtest.net/result/c/e35b325b-2a8f-4258-8db2-1a5a874f63a0

Then test with waveform.com bufferbloat test -- A+
https://www.waveform.com/tools/bufferbloat?test-id=da64d035-23be-4e6a-a54e-39d0d9b28d47

2) Ubuntu 22.04 VM behind behind OpenWRT virtual router 1 (A+)
First test to show that the SQM rule works fine.

Bash:

mcuee@ubuntu2204vmbr0:~/build/speedtest$ ./speedtest -s 13623

   Speedtest by Ookla

      Server: Singtel - Singapore (id: 13623)
         ISP: Singtel Fibre
Idle Latency:     1.72 ms   (jitter: 1.36ms, low: 1.43ms, high: 4.51ms)
    Download:   813.52 Mbps (data used: 599.2 MB)                                                   
                  3.41 ms   (jitter: 1.68ms, low: 1.31ms, high: 13.24ms)
      Upload:   190.09 Mbps (data used: 231.5 MB)                                                   
                  1.39 ms   (jitter: 0.69ms, low: 1.10ms, high: 12.86ms)
 Packet Loss:     0.0%
  Result URL: https://www.speedtest.net/result/c/3e803254-0517-4bbf-8767-58259afc38b6

Then test with waveform.com bufferbloat test -- A+
https://www.waveform.com/tools/bufferbloat?test-id=79c2d004-8190-4840-9fe2-aeb4d125e174

xiaofan · Mar 30, 2024

However, waveform.com speed is on the low side so sometimes we will have doubts on the validity of the test results.

So I use crusader to test again with my Acer Windows 11 laptop and Ugreen USB 3 to 2.5G adapter, behind the OpenWRT virtual router 1 (2.5G LAN).

Without SQM:

With SQM (CAKE)

xiaofan · Mar 31, 2024

hwzlite said:
Yes, he is the one who co-founded the Bufferbloat Project , the birth of FQ-CoDel, Cake SQM

One of his presentations which you mentioned earlier.
https://blog.apnic.net/2020/01/22/bufferbloat-may-be-solved-but-its-not-over-yet/

xiaofan · Mar 31, 2024

Discussion of why it is difficult to get an off-the-shelf low cost (about US$100) wireless router which can work with OpenWRT (easy installation by new users) and support SQM >= 300Mbps.

https://forum.openwrt.org/t/best-newcomer-routers-2024/189050/36

From Rich Brown.
The following routers meet (or are close to) the criteria, and links to their recommendations

GL.iNet GL-MT3000 44 - Right around $100, available now, easy installation, plenty of horsepower - Thanks @qunvureze for the recommendation
Netgear R7800 16 - Older model, well under $100 used, easy to flash, only handles SQM to about 180mbps
GL.iNet GL-MT6000 41 - This appears to be a great router, and is now officially supported with OpenWrt 23.05.3. Its price (~US$159, March 2024) doesn't quite meet the criteria, so you will need to decide if it's worth the extra money.

Honorable mentions: These seem to be good routers, but don't meet all the criteria for newcomers:

Netgear WAX206 - Looks easy to flash; similar power as Linksys E8450; may be available under US$100 on eBay
Cudy WR3000 - Two-step install process is confusing for newcomers; reported to max out at 360-380Mbps with SQM 4; under $60 and otherwise looks interesting
TP-Link AX23 - £53.99 on Amazon UK, likely can only support SQM at 100Mbps; nice link to the Device Page instructions
~~Linksys E8450/Belkin RT3200 - lots of power, highly-regarded, but has a confusing two-step install process for newcomers.~~ No longer recommended for newcomers. This topic 6 describes difficulties with flashing/recovering/bricking

xiaofan · Mar 31, 2024

The post from slh is more in line -- using x86_64 mini PC (like Intel N100) and then dump AP.

Newer WiFi 6 routers may get better support by OpenWRT as well which will do SQM>=300Mbps but the low cost ones may still struggle with SQM at around 1Gbps.

+++++++++++++++++++
Even the first generation of wifi6 devices (e.g. mt7622bv+mt7915) is getting replaced by the (cost- and feature) optimized second wave (e.g. filogic 820/ 830). What you end up with, are either cheap low-end devices (e.g. mt7621a+mt7915DBDC) which have no chance of meeting your performance requirements - or devices that are simply too new for your 23.05.2 support requirement (and yes, also exceed your price cap and often the ease-of-flashing one as well).

+++++++++++++++++
Btw., the most you may expect from the r7800/ xr500 with sqm/cake would be around 180 MBit/s, less if PPPoE or similar things are involved.

The GL.Inet GL-MT6000 is probably most promising (I have no personal experiences with it, GL.Inet or filogic 820/ 830 in general though), but it's obviously outside of your price range.

The OpenWrt One should come close as well (claiming to meet all of your strict requirements, aside from not being available yet), but the 1+1 ethernet ports are a deterrence for many normal users (including myself, I've made the experience that less than 1+3 or better 1+4 ports on the router does severely limit my options).

Things like the NanoPi r2s/ r4s or a four-port x86_64 system (e.g. alderlake-n/ n100) solve the sqm/cake vs performance issue easily, but obviously lack the wireless capabilites (but maybe the existing/ old router (with its OEM firmware) can be used as dumb-AP, in the worst case in a double-NAT setup).
+++++++++++++++++

xiaofan · Mar 31, 2024

Performance of GL.iNET Flint 2 (GL-MT6000, MediaTek Filogic 830 CPU) with regard to SQM.
https://forum.openwrt.org/t/gl-inet-flint-2-gl-mt6000-discussions/173524/262

Yea hardware flow offloading works great on this target, I get similar results and speeds. Meanwhile SQM does not seem as consistent on Filogic 830. Shockingly my 7 year old WRT32X (mvebu) has better performance with cake (as long as nothing else is being done on the router, be in Samba, WiFi, etc. where it becomes CPU limited).

here you go, ran two tests with SQM fq_codel and Bittorent. One at 800/20 Mbps cap (hits ~750 Mbps if you add up the speeds shown in the pic), one at 600/15 Mbps cap (achieves a semi A+ rating, hits ~550 Mbps if you add up the speeds in pic)... but the ping spread is not ideal. Cake is simply a better algorithm for bufferbloat, but I'm not getting these speeds with that. Overall results aren't too bad, but I prefer the HFO+WED approach on this target.

I'm hoping kernel 6.1 improves things or maybe we find some irqbalance/affinity tweaks.

xiaofan · Mar 31, 2024

1) Performance of Asus TUF-AX6000 (MediaTek Filogic 830 CPU) SQM when running OpenWRT.
https://forum.openwrt.org/t/asus-tuf-gaming-ax6000-support/163002/189?page=10

SQM does not work with hardware offloading at all.
SQM works just fine when only software offloading enabled (no hardware) and when packet steering is enabled (in Network - Interfaces - Global network options). - CPU load is significant, and it's up to 60% on all cores, but SQM works perfectly.

2) Similar discussion for Asus TUF-AX4200 (MediaTek Filogic 830 CPU) when running OpenWRT
https://forum.openwrt.org/t/asus-tuf-ax4200-support/155738/376?page=19

without SQM:
Grade B
DL 910Mbps ; UL 402 Mbps ; ping 12ms +30DL +21UL.

With SQM cake/piece_of_cake (limit at 845000 / 360000)
grade A
DL 699Mbps ; UL 310 MBps ; ping 14ms +27DL +0UL

With SQM FQ_codel/simplest_tbf (limit at 845000 / 360000)
grade A+
DL 826.0 Mbps ; UL 328.9 Mbps ; ping 13ms +0DL +0UL

With SQM FQ_codel/simple (limit at 900000 / 390000)
grade A+
DL 880.2 Mbps ; UL 388.8 Mbps ; ping 10ms +2DL +0UL
simplest.qos and simplest_tbf.qos give the same results as simple.

xiaofan · Mar 31, 2024

Interesting read as well for StarLink. Rev 4 is much better than Rev 3 in terms of speed and latency.
https://olegkutkov.me/2024/02/12/starlink-terminal-revision-4-overview-and-tests/

hwzlite · Mar 31, 2024

xiaofan said:
Interesting read as well for StarLink. Rev 4 is much better than Rev 3 in terms of speed and latency.
https://olegkutkov.me/2024/02/12/starlink-terminal-revision-4-overview-and-tests/

Wow...love teardown porn, kudo the OP for the "housing is quite hard to disassemble without partial destruction."

Guess another good use case for cake-autorate.

hwzlite · Mar 31, 2024

xiaofan said:
In the previous two wired test cases, OpenWRT virtual router 1 gets A.

How to get A+? Limit the upload speed to 200Mbps is one potential solution judging from the test results.

Using CAKE in the following tests.

1) Acer Windows 11 laptop with Ugreen USB 3 to 2.5G adapter, behind OpenWRT virtual router 1 (A+)

First test to show that the SQM rule works fine.

Bash:

PS C:\work\speedtest\ookla-speedtest-1.2.0-win64> .\speedtest.exe -s 13623 Speedtest by Ookla Server: Singtel - Singapore (id: 13623) ISP: Singtel Fibre Idle Latency: 2.58 ms (jitter: 0.31ms, low: 2.15ms, high: 2.87ms) Download: 920.84 Mbps (data used: 450.2 MB) 3.55 ms (jitter: 0.64ms, low: 1.88ms, high: 5.04ms) Upload: 192.71 Mbps (data used: 161.5 MB) 1.96 ms (jitter: 0.51ms, low: 1.64ms, high: 24.64ms) Packet Loss: 0.0% Result URL: https://www.speedtest.net/result/c/e35b325b-2a8f-4258-8db2-1a5a874f63a0

Then test with waveform.com bufferbloat test -- A+
https://www.waveform.com/tools/bufferbloat?test-id=da64d035-23be-4e6a-a54e-39d0d9b28d47

2) Ubuntu 22.04 VM behind behind OpenWRT virtual router 1 (A+)
First test to show that the SQM rule works fine.

Bash:

mcuee@ubuntu2204vmbr0:~/build/speedtest$ ./speedtest -s 13623 Speedtest by Ookla Server: Singtel - Singapore (id: 13623) ISP: Singtel Fibre Idle Latency: 1.72 ms (jitter: 1.36ms, low: 1.43ms, high: 4.51ms) Download: 813.52 Mbps (data used: 599.2 MB) 3.41 ms (jitter: 1.68ms, low: 1.31ms, high: 13.24ms) Upload: 190.09 Mbps (data used: 231.5 MB) 1.39 ms (jitter: 0.69ms, low: 1.10ms, high: 12.86ms) Packet Loss: 0.0% Result URL: https://www.speedtest.net/result/c/3e803254-0517-4bbf-8767-58259afc38b6

Then test with waveform.com bufferbloat test -- A+
https://www.waveform.com/tools/bufferbloat?test-id=79c2d004-8190-4840-9fe2-aeb4d125e174

Btw download throughput can be potentially improved by Adding ack-filtering to cake for test cases that limit upload speed.

Set under OpenWRT > SQM > Queue Discipline > Advanced Configuration > Dangerous Configuration > Qdisc options (egress) > add "ack-filter" setting

xiaofan · Mar 31, 2024

Try to see if Intel N100 can do 2.5Gbps SQM or not -- looks like no issues.

The following tests are not done through the internet as I only have 1Gbps Singtel Fibre. The following tests are doing using internal home network using crusader.

Server: Ubuntu 22.04 LXC container (192.168.50.15) on Intel N100 mini PC running Proxmox PVE 8.0.
Client: Windows 11 laptop with Ugreen USB 3 to 2.5G adapter, connected to OpenWRT virtual router 2.5G LAN port.
OpenWRT 23.05 virtual router on Intel N100 mini PC running Proxmox PVE 8.0 (Double NAT behind Asus RT-AX86U).
OpenWRT 23.05 virtual router WAN -- 192.168.50.134
OpenWRT 23.05 virtual rotuer LAN -- 192.168.48.1

Typical download speed 2.37Gbps and upload speed 2.12Gbps without SQM rules, tested using iperf3.

I use iperf3 to make sure the SQM rules work fine before carrying out the bufferbloat tests using crusader.

1) No SQM -- upload latency gets higher

2) Using SQM, cake, disable ingress shaping, upload limit 1.2Gbps

3) Using SQM, cake, disable ingress shaping, upload limit 1.8 Gbps, somehow download latency gets affected

4) Using SQM, cake, download limit 2Gbps, upload limit 2 Gbps, both download and upload seem to work fine.

xiaofan · Mar 31, 2024

hwzlite said:
Btw download throughput can be potentially improved by Adding ack-filtering to cake for test cases that limit upload speed.

Set under OpenWRT > SQM > Queue Discipline > Advanced Configuration > Dangerous Configuration > Qdisc options (egress) > add "ack-filter" setting

It is a bit tricky to get good Waveform.com speed.

That is why now I will trust my own Internet and Intranet test results using crusader more than waveform.com bufferbloat test.

1. Behind OpenWRT virtual router 1 (Single NAT)

Waveform.com speed is rather low so I have doubts on the validity of the tests.

Singtel ONT -- TL-SG105E smart switch -- OpenWRT virtual router 1 WAN (2.5G port)
OpenWRT virtual router 2.5G LAN -- Acer Windows 11 laptop with Ugreen USB to 2.5G adapter

1) Without SQM
Score A, download 432.6 Mbps, upload 244.0 Mbps
https://www.waveform.com/tools/bufferbloat?test-id=89954945-b802-45d8-ba91-368a57d2b1b0

2) With SQM, cake, disable ingress shaping, upload limit 200Mbps
Score A+, download 306.7 Mbps, upload 141.8 Mbps
https://www.waveform.com/tools/bufferbloat?test-id=0b587670-166f-413c-b900-741249bf6261

3) With SQM, cake, disable ingress shaping, upload limit 200Mbps, adding the "ack-filter" setting to both ingress and egress.
Score A+, download 454.3 Mbps, upload 131.2 Mbps
https://www.waveform.com/tools/bufferbloat?test-id=35d0829f-2710-49bc-8f8b-481e3a217789

2. Beind OpenWRT virtual router 2 which is behind Asus RT-AX86U (Double NAT)

Speed from Waveform.com is faster than Case 1.

Singtel ONT -- TL-SG105E smart switch -- Asus RT-AX86U WAN
RT-AX86U 2.5G LAN -- OpenWRT virtual router 2 WAN (2.5G port)
OpenWRT virtual router 2 2.5G LAN -- Acer Windows 11 laptop with Ugreen USB to 2.5G adapter

1) Without SQM: tested twice, both A+, download is about 660 Mbps, upload > 500Mbps
https://www.waveform.com/tools/bufferbloat?test-id=5ad758ce-831e-4fa9-b1ab-ed9089e6b267
https://www.waveform.com/tools/bufferbloat?test-id=7bda7190-9b33-4e48-9dce-c26aea9fe011

2) With SQM, cake, disable ingress shaping, upload limit 200Mbps
Score A, download 636.4Mbps, upload 194.8Mbps
https://www.waveform.com/tools/bufferbloat?test-id=1a966cc5-dc92-4ade-9ca3-18e0cc3f5aae

xiaofan · Mar 31, 2024

Initial testing of cake-autorate. I am not so sure how effective it is versus the normal SQM rules.

Installation instruction:
https://github.com/lynxthecat/cake-autorate/blob/master/INSTALLATION.md

I feel it will be more useful for wireless so the following testing are donw with wireless.
Singtel ONT -- TL-SG105E smart switch -- OpenWRT virtual router 1 WAN (2.5G port)
OpenWRT virtual router 1 LAN (2.5G port) --Asus RT-AX82U AP --wireless -- Acer Windows 11 laptop using Intel AX201WiFi 6 adapter

1) Without SQM -- B and C
https://www.waveform.com/tools/bufferbloat?test-id=85c5b897-3739-47db-937b-2a207f1f683c
https://www.waveform.com/tools/bufferbloat?test-id=3ece9119-0b64-453d-afb8-53f62a4e38c6

OOKla SpeedTest results for reference:

Bash:

PS C:\work\speedtest\ookla-speedtest-1.2.0-win64> .\speedtest.exe -s 13623

   Speedtest by Ookla

      Server: Singtel - Singapore (id: 13623)
         ISP: Singtel Fibre
Idle Latency:     4.88 ms   (jitter: 1.01ms, low: 3.81ms, high: 7.26ms)
    Download:   718.73 Mbps (data used: 513.4 MB)
                 11.67 ms   (jitter: 2.67ms, low: 7.72ms, high: 36.95ms)
      Upload:   479.13 Mbps (data used: 489.9 MB)
                 14.25 ms   (jitter: 4.84ms, low: 3.29ms, high: 36.15ms)
 Packet Loss:     0.0%
  Result URL: https://www.speedtest.net/result/c/3935ceb1-0c3e-420b-8980-e34d6c8e25bd

Bash:

PS C:\work\speedtest\ookla-speedtest-1.2.0-win64> .\speedtest.exe -s 13623

   Speedtest by Ookla

      Server: Singtel - Singapore (id: 13623)
         ISP: Singtel Fibre
Idle Latency:     3.87 ms   (jitter: 0.16ms, low: 3.64ms, high: 4.29ms)
    Download:   656.87 Mbps (data used: 657.1 MB)
                 11.18 ms   (jitter: 2.67ms, low: 3.85ms, high: 20.18ms)
      Upload:   351.67 Mbps (data used: 607.1 MB)
                  6.51 ms   (jitter: 2.64ms, low: 3.34ms, high: 24.28ms)
 Packet Loss:     0.0%
  Result URL: https://www.speedtest.net/result/c/5d5d78af-a130-402c-96a8-965dbfb54d87

Configuration

Bash:

### For multihomed setups, it is the responsibility of the user to ensure that the probes
### sent by this instance of cake-autorate actually travel through these interfaces.
### See ping_extra_args and ping_prefix_string

dl_if=ifb4eth0 # download interface
ul_if=eth0    # upload interface

# Set either of the below to 0 to adjust one direction only
# or alternatively set both to 0 to simply use cake-autorate to monitor a connection
adjust_dl_shaper_rate=1 # enable (1) or disable (0) actually changing the dl shaper rate
adjust_ul_shaper_rate=1 # enable (1) or disable (0) actually changing the ul shaper rate

min_dl_shaper_rate_kbps=5000  # minimum bandwidth for download (Kbit/s)
base_dl_shaper_rate_kbps=650000 # steady state bandwidth for download (Kbit/s)
max_dl_shaper_rate_kbps=750000  # maximum bandwidth for download (Kbit/s)

min_ul_shaper_rate_kbps=5000  # minimum bandwidth for upload (Kbit/s)
base_ul_shaper_rate_kbps=200000 # steady state bandwidth for upload (KBit/s)
max_ul_shaper_rate_kbps=380000  # maximum bandwidth for upload (Kbit/s)

connection_active_thr_kbps=20  # threshold in Kbit/s below which dl/ul is considered idle

xiaofan · Mar 31, 2024

More test report of cake-autorate here. I am not so sure I am doing the things correctly here.
https://github.com/lynxthecat/cake-autorate/issues/274

xiaofan · Mar 31, 2024

Need to read more about cake with adaptive bandwidth here.

FYI/A: Bufferbloat 101

High Supremacy Member

Master Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

Master Member

Master Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

High Supremacy Member

CAKE w/ Adaptive Bandwidth (new thread)​

CAKE w/ Adaptive Bandwidth (new thread)