CAKE w/ Adaptive Bandwidth

moeller0 · 2024-10-14T06:14:11.989Z

Amalvajdan:

on that note, is it possible that cake-autorate continuously pinging the reflectors are giving me better upload performance?

Possible for sure, but likely I have no idea...
Mobile data transmissions are scheduled conceptually by the base station, that is user equipment like your modem request transmission slots over a narrow "request channel" and the base station grants these over the same channel. Data then is transmitted only within the allotted time-frequency bands. I have no clue whether a UE only asks for transmit slots or whether it also tells the base station about the queued up data, but my hunch is the base station has an idea about the queued data. Others on this forum know much more about the details that I stop speculating here...
Note how both the user equipment and the basestation can influence this process by requesting with less or more data in the user side buffers and the bade station can also grant the time-frequency slots earlier or later.
Airtime slots likely for efficiency reasons have a minimum (or even a fixed) length, let's arbitrarily assume 1500 bytes for our discussion. Say your game sends a 500 byte podition/action update packet, this hits the modem. The modem now might bevtempted to not request a transmission slot immediately but wait a bit on the chance that some more data might arrive in the near future allowing it to fill up that 1500 byte minimum and hence use the scarce resource airtime more efficiently... If you then consider that for capacity tests (often called speed tests, but data volume over time is not speed, that would be distance over time) efficient use of airtime is quite important as wasted airtime reduces the achievable throughput, while a few dozend millisecond delay/jitter do not meaningfully diminish capacity tests.
That IMHO would incentivise an ISP to configure their mobile networks for throughput over low latency.
This is why in theory a constant low load might be usefull in keeping a steady flow of timeslots coming to your modem, which then can be used for game packets. But all of this sits on top of unsubstantiated assumptions so might be completely wrong for your situation.

Amalvajdan · 2024-10-14T13:50:40.167Z

moeller0:

Maybe your ISP does IPv6 only

Hi moeller, i think my ISP supports ipv4 only. fping -D -e -4 -b 1450 -i 1 -p 10 -l gstatic.com 9.9.9.9 is not working, both servers are returning 100% packet loss.
do you have any further advice for me? i'm willing to try anything

Lynx · 2024-10-14T13:51:52.195Z

opkg install iputils-ping
ping -i 0.010 -s 1464 8.8.8.8

moeller0 · 2024-10-14T16:26:54.611Z

The packet loss is fine, as long as the packets are sent to the servers in the first place, here we really only want to convince your modem to request transmit slots in a timely fashion, so as long as it is not the modem that drops these packets 100% packet loss is fine (as long as these two are not in your set of reflectors ).

moeller0 · 2024-10-14T16:28:41.154Z

The idea was to slam two packets into the modem asap and the wait for 10ms for the next two, so the modem would see ~3kB every 10ms, hopefully enough to request a transmit slot soon.

Lynx · 2024-10-14T16:38:35.159Z

Out of curiosity does @Amalvajdan experience:

CAKE w/ Adaptive Bandwidth [August 2022 to March 2024]

On my connection I am seeing a noticeable increase in upload one way delay on saturating download, whereas on saturating upload there is not a similar increase in download one way delay.

Amalvajdan · 2024-10-14T18:26:16.065Z

while saturating download: https://www.waveform.com/tools/bufferbloat?test-id=b858d1cd-6d4a-4718-b40b-da423b3e39a9 (did a couple of tests, but only this one had a reliable unloaded value) cpu usage was 99%
saturated upload: https://www.waveform.com/tools/bufferbloat?test-id=71f95970-7714-41db-b00b-0c685b334274 (except for last few seconds of the test)

tbh my unloaded ping goes up with any significant network activity, so i'm not sure how realiable this results are.

Lynx · 2024-10-14T19:13:55.349Z

Could you run a waveform test, then repeat with running:

ping -i 0.010 -s 1464 8.8.8.8

at the same time?

dlakelan · 2024-10-14T19:40:30.752Z

moeller0:

~3kB every 10ms, hopefully enough to request a transmit slot soon.

Yes, but also ~2.5Mbps of flow so maybe enough that it start to interfere with the real data.

moeller0 · 2024-10-14T20:29:41.334Z

Sure this is primarily for testing/exploration, not a long term solution.

dlakelan · 2024-10-14T22:34:30.357Z

Yes, of course, but even for testing might want to try a variety of smaller packet sizes to see if the results depend on the bandwidth used.

Lynx · 2024-10-14T22:54:51.355Z

Is there a protocol for dealing with packet loss in which redundancy is added? Such a protocol might have use here? Not because we’re dealing with packet loss, but rather because the redundancy padding might whip the sleepy base station scheduler into shape?

moeller0 · 2024-10-15T05:57:28.151Z

Doesn't work, my hypothesis is that the modem requests transmit slots fadter if it has same queued data, so to queue 2 full packets, we need ~3kB in an interval small eough to be useful for his gaming traffic...
Again, this is a hypothesis only and might just be wrong...

moeller0 · 2024-10-15T05:59:00.463Z

Some protocols allow forward error correction or redundancy, but unless both sides speak such a protocol it is not going to be useful I fear.

Amalvajdan · 2024-10-29T05:54:43.496Z

Hi Moeller0,

this was an interesting hypothesis and i think you maybe right.

i bought a second router (HTC 5G HUB, basically an android phone with ethernet and USB 3.0) and placed it outside my window and the upload latency basically disappeared.
I put my nokia router in the same exact spot (fiddled around with orientation to get best signal) but the upload latency came back.

although the upload latency/ jitter went away, the HTC seems to be a weaker device compared to my Nokia, so i have to choose between consistent but more upload jitter (nokia) or no upload jitter but occasional spikes (HTC). It is what it is is

pstlr78 · 2024-11-05T19:10:51.134Z

@Lynx is there any luci interface for cake-autorate? I understand that all configuration (service installation, dl_if, ul_if, etc.) needs to be set via SSH commands (not in luci), is that correct? Thanks

Lynx · 2024-11-05T23:43:01.295Z

Yes that’s right - just config file at the moment. We’d need one or more volunteers to code up a LuCi interface and help package up cake-autorate.

famewolf · 2024-11-27T05:52:05.254Z

Anyone have apks for the app instead of opkg which has been abandoned in snapshots?

Lynx · 2024-11-27T09:30:01.887Z

There never was a package for this service-based script. But the GitHub repository has been updated in respect of the installation procedure for future compatibility.

Gigahertz7971 · 2024-11-28T03:08:58.370Z

I need some help with the values for min, max and base bandwidth values.

So my ISP throttles both download and upload speeds to 5Mbps when we exceed a certain usage limit. If I read the GitHub readme correctly should I then set the min bandwidth to 5Mbps or a little lower? As for max bandwidth should I set it to the the speeds I get before hitting the usage limit? And for the base bandwidth should I set it to the value of min bandwidth or something similar to it?