Reducing multiplexing latencies still further in wifi

Spacebar · June 3, 2024, 8:00am

I have this patch.

---
 net/mac80211/sta_info.c | 10 +---------
 net/mac80211/tx.c       |  4 ++--
 2 files changed, 3 insertions(+), 11 deletions(-)

--- a/net/mac80211/sta_info.c
+++ b/net/mac80211/sta_info.c
@@ -2893,15 +2893,7 @@ static void sta_update_codel_params(stru
 	if (ieee80211_hw_check(&sta->sdata->local->hw, HAS_TX_QUEUE))
 		return;

-	if (thr && thr < STA_SLOW_THRESHOLD * sta->local->num_sta) {
-		sta->cparams.target = MS2TIME(50);
-		sta->cparams.interval = MS2TIME(300);
-		sta->cparams.ecn = false;
-	} else {
-		sta->cparams.target = MS2TIME(20);
-		sta->cparams.interval = MS2TIME(100);
-		sta->cparams.ecn = true;
-	}
+	return;
 }

 void ieee80211_sta_set_expected_throughput(struct ieee80211_sta *pubsta,
--- a/net/mac80211/tx.c
+++ b/net/mac80211/tx.c
@@ -1610,8 +1610,8 @@ int ieee80211_txq_setup_flows(struct iee
 		fq->memory_limit = 4 << 20; /* 4 Mbytes */

 	codel_params_init(&local->cparams);
-	local->cparams.interval = MS2TIME(100);
-	local->cparams.target = MS2TIME(20);
+	local->cparams.interval = MS2TIME(40);
+	local->cparams.target = MS2TIME(3);
 	local->cparams.ecn = true;

 	local->cvars = kcalloc(fq->flows_cnt, sizeof(local->cvars[0]),

And if it helps, here is a test with crusader
wax620: OpenWrt support for Netgear WAX620 - #81 by Spacebar
wax630: Add support for Netgear WAX630 - #48 by Spacebar

amteza · June 4, 2024, 6:51pm

3 ms and 40 ms, how much bandwidth do you lose?

Spacebar · June 4, 2024, 7:15pm

Tx rate is 1170 and bandwidth is 940-ish. So that would be 80% ish... I loose 20%.

amteza · June 5, 2024, 2:20am

Are you touching any other parameters? AQL?

Spacebar · June 5, 2024, 4:53am

IPQ8074 does not support AQL. But only the parameters presented in Luci.
I'm also downgrading my WLAN.HK firmware to 2.7 because I saw that i.e version 2.9 the latency doubled. But need more data to verify that...

Note: I'm also using AC mode, not AX. Because when I was testing speed with my iPhone the latency went all yo-yo compared to AC mode that is stable.

castiel652 · June 5, 2024, 8:40am

There was a patch to add it but they never resend it.
You can try it yourself if you are feeling adventurous.
https://patchwork.kernel.org/project/linux-wireless/patch/20230501130725.7171-1-quic_tamizhr@quicinc.com/

Spacebar · June 5, 2024, 8:44am

Ooooh that's nice. Yes, this is something I would love to test. Thanks

Edit: Well, it did build and run for a short time before it crashed.

Supported extended features:
* [ AQL ]: Airtime Queue Limits (AQL)

But yeah, didn't go so well. But fun to give it a go.

_FailSafe · June 5, 2024, 2:50pm

@pesa1234 has been working on a nice build for the GL-MT6000 device. I have been helping to test these settings. The latest build is 8ms and 80ms and here are some benchmarks I ran with those values:

However, prior to testing with @pesa1234's builds, I had been running with 5ms and 50ms for a LONG time.

Spacebar · June 5, 2024, 3:01pm

My vote goes for custom values. If not I think 5/50.

If the target is too low, how are we punished? Bandwidth?

castiel652 · June 5, 2024, 4:12pm

Do you have the crash log?

Spacebar · June 5, 2024, 5:09pm

Hope you understand more than me

[   36.147634] qcom-q6v5-wcss-pil cd00000.q6v5_wcss: fatal error received:
[   36.147634] QC Image Version: QC_IMAGE_VERSION_STRING=WLAN.HK.2.7.0.1-01744-QCAHKSWPL_SILICONZ-1
[   36.147634] Image Variant : IMAGE_VARIANT_STRING=8074.wlanfw.eval_v2Q
[   36.147634]
[   36.147634]     :Excep  :0 Exception detectedparam0 :zero, param1 :zero, param2 :zero.
[   36.147634] Thread ID      : 0x00000069  Thread name    : WLAN RT0  Process ID     : 0
[   36.147634] Register:
[   36.147634] SP : 0x4bfacea0
[   36.147634] FP : 0x4bfaceb0
[   36.147634] PC : 0x4b15be0c
[   36.147634] SSR : 0x00000003
[   36.147634] BADVA : 0x0080cca4
[   36.147634] LR : 0x4b15be14
[   36.147634]
[   36.147634] Stack Dump
[   36.147634] from : 0x4bfacea0
[   36.147634] to   : 0x4bfad400
[   36.147634]
[   36.194179] remoteproc remoteproc0: crash detected in cd00000.q6v5_wcss: type fatal error
[   36.216420] remoteproc remoteproc0: handling crash #1 in cd00000.q6v5_wcss
[   36.224569] remoteproc remoteproc0: recovering cd00000.q6v5_wcss
[   36.257000] remoteproc remoteproc0: stopped remote processor cd00000.q6v5_wcss
[   36.257224] ath11k c000000.wifi: failed to clear rx_filter for monitor status ring: (-108)
[   36.596553] remoteproc remoteproc0: remote processor cd00000.q6v5_wcss is now up
[   36.640495] ath11k c000000.wifi: qmi ignore invalid mem req type 3
[   36.648005] ath11k c000000.wifi: chip_id 0x0 chip_family 0x0 board_id 0xff soc_id 0xffffffff
[   36.648059] ath11k c000000.wifi: fw_version 0x270204a5 fw_build_timestamp 2022-08-04 13:05 fw_build_id WLAN.HK.2.7.0.1-01744-QCAHKSWPL_SILICONZ-1
[   36.664964] ath11k c000000.wifi: Last interrupt received for each CE:
[   36.668484] ath11k c000000.wifi: CE_id 0 pipe_num 0 21880ms before
[   36.674914] ath11k c000000.wifi: CE_id 1 pipe_num 1 830ms before
[   36.680955] ath11k c000000.wifi: CE_id 2 pipe_num 2 830ms before
[   36.687132] ath11k c000000.wifi: CE_id 3 pipe_num 3 840ms before
[   36.693127] ath11k c000000.wifi: CE_id 5 pipe_num 5 4294703996ms before
[   36.699100] ath11k c000000.wifi: CE_id 7 pipe_num 7 1160ms before
[   36.705450] ath11k c000000.wifi: CE_id 9 pipe_num 9 4294704006ms before
[   36.711692] ath11k c000000.wifi: CE_id 10 pipe_num 10 4294704016ms before
[   36.718109] ath11k c000000.wifi: CE_id 11 pipe_num 11 4294704016ms before
[   36.725065] ath11k c000000.wifi:
[   36.725065] Last interrupt received for each group:
[   36.731834] ath11k c000000.wifi: group_id 0 21680ms before
[   36.739982] ath11k c000000.wifi: group_id 1 21680ms before
[   36.745372] ath11k c000000.wifi: group_id 2 4294704046ms before
[   36.750833] ath11k c000000.wifi: group_id 3 4294704046ms before
[   36.756657] ath11k c000000.wifi: group_id 4 1000ms before
[   36.762554] ath11k c000000.wifi: group_id 5 2280ms before
[   36.768106] ath11k c000000.wifi: group_id 6 4294704066ms before
[   36.773493] ath11k c000000.wifi: group_id 7 4294704076ms before
[   36.779218] ath11k c000000.wifi: group_id 8 4294704076ms before
[   36.785126] ath11k c000000.wifi: group_id 9 4294704086ms before
[   36.791108] ath11k c000000.wifi: group_id 10 4294704096ms before
[   36.796928] ath11k c000000.wifi: dst srng id 0 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704096ms
[   36.803219] ath11k c000000.wifi: dst srng id 1 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704106ms
[   36.813699] ath11k c000000.wifi: dst srng id 2 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704116ms
[   36.824459] ath11k c000000.wifi: dst srng id 3 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704126ms
[   36.835223] ath11k c000000.wifi: dst srng id 4 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 21780ms
[   36.845988] ath11k c000000.wifi: src srng id 5 hp 0, reap_hp 248, cur tp 0, cached tp 0 last tp 0 napi processed before 4294704146ms
[   36.856406] ath11k c000000.wifi: src srng id 8 hp 0, reap_hp 2550, cur tp 0, cached tp 0 last tp 0 napi processed before 4294704156ms
[   36.868471] ath11k c000000.wifi: dst srng id 9 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704166ms
[   36.880362] ath11k c000000.wifi: src srng id 16 hp 0, reap_hp 4088, cur tp 0, cached tp 0 last tp 0 napi processed before 4294704176ms
[   36.890969] ath11k c000000.wifi: src srng id 17 hp 0, reap_hp 4088, cur tp 0, cached tp 0 last tp 0 napi processed before 4294704186ms
[   36.903132] ath11k c000000.wifi: src srng id 18 hp 0, reap_hp 4088, cur tp 0, cached tp 0 last tp 0 napi processed before 4294704206ms
[   36.915175] ath11k c000000.wifi: src srng id 24 hp 0, reap_hp 248, cur tp 0, cached tp 0 last tp 0 napi processed before 4294704216ms
[   36.927239] ath11k c000000.wifi: dst srng id 25 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704226ms
[   36.939308] ath11k c000000.wifi: src srng id 32 hp 20, reap_hp 16, cur tp 20, cached tp 20 last tp 16 napi processed before 22150ms
[   36.950243] ath11k c000000.wifi: src srng id 35 hp 16, reap_hp 12, cur tp 16, cached tp 16 last tp 12 napi processed before 1100ms
[   36.961787] ath11k c000000.wifi: src srng id 36 hp 72, reap_hp 68, cur tp 72, cached tp 72 last tp 68 napi processed before 1430ms
[   36.973592] ath11k c000000.wifi: src srng id 39 hp 64, reap_hp 60, cur tp 64, cached tp 64 last tp 60 napi processed before 1440ms
[   36.985324] ath11k c000000.wifi: src srng id 41 hp 0, reap_hp 124, cur tp 0, cached tp 0 last tp 0 napi processed before 4294704286ms
[   36.997046] ath11k c000000.wifi: src srng id 57 hp 712, reap_hp 712, cur tp 716, cached tp 716 last tp 716 napi processed before 1150ms
[   37.009097] ath11k c000000.wifi: src srng id 58 hp 286, reap_hp 286, cur tp 290, cached tp 290 last tp 290 napi processed before 1160ms
[   37.021076] ath11k c000000.wifi: src srng id 61 hp 1020, reap_hp 1020, cur tp 0, cached tp 0 last tp 0 napi processed before 22240ms
[   37.033229] ath11k c000000.wifi: src srng id 66 hp 1020, reap_hp 1020, cur tp 0, cached tp 0 last tp 0 napi processed before 22250ms
[   37.045382] ath11k c000000.wifi: dst srng id 81 tp 1432, cur hp 1432, cached hp 1432 last hp 1432 napi processed before 1200ms
[   37.057276] ath11k c000000.wifi: dst srng id 82 tp 580, cur hp 580, cached hp 580 last hp 580 napi processed before 1210ms
[   37.068469] ath11k c000000.wifi: dst srng id 85 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704366ms
[   37.079495] ath11k c000000.wifi: dst srng id 90 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704376ms
[   37.090519] ath11k c000000.wifi: src srng id 104 hp 65532, reap_hp 65532, cur tp 0, cached tp 0 last tp 0 napi processed before 22310ms
[   37.101371] ath11k c000000.wifi: src srng id 105 hp 0, reap_hp 504, cur tp 0, cached tp 0 last tp 0 napi processed before 4294704406ms
[   37.113266] ath11k c000000.wifi: dst srng id 106 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 22060ms
[   37.125413] ath11k c000000.wifi: dst srng id 107 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 22070ms
[   37.136091] ath11k c000000.wifi: dst srng id 108 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704436ms
[   37.146596] ath11k c000000.wifi: dst srng id 109 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 22090ms
[   37.157448] ath11k c000000.wifi: src srng id 128 hp 8190, reap_hp 8190, cur tp 0, cached tp 0 last tp 0 napi processed before 22100ms
[   37.168037] ath11k c000000.wifi: src srng id 130 hp 8190, reap_hp 8190, cur tp 0, cached tp 0 last tp 0 napi processed before 22140ms
[   37.179933] ath11k c000000.wifi: src srng id 132 hp 242, reap_hp 242, cur tp 244, cached tp 244 last tp 244 napi processed before 1420ms
[   37.191911] ath11k c000000.wifi: dst srng id 133 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 22140ms
[   37.204233] ath11k c000000.wifi: dst srng id 134 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704506ms
[   37.214736] ath11k c000000.wifi: src srng id 135 hp 8190, reap_hp 8190, cur tp 0, cached tp 0 last tp 0 napi processed before 22190ms
[   37.225590] ath11k c000000.wifi: src srng id 143 hp 8190, reap_hp 8190, cur tp 0, cached tp 0 last tp 0 napi processed before 22170ms
[   37.237569] ath11k c000000.wifi: src srng id 145 hp 8190, reap_hp 8190, cur tp 0, cached tp 0 last tp 0 napi processed before 22180ms
[   37.249547] ath11k c000000.wifi: src srng id 147 hp 14, reap_hp 14, cur tp 16, cached tp 16 last tp 16 napi processed before 2760ms
[   37.261540] ath11k c000000.wifi: dst srng id 148 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 22210ms
[   37.273189] ath11k c000000.wifi: dst srng id 149 tp 0, cur hp 0, cached hp 0 last hp 0 napi processed before 4294704576ms
[   37.283922] ath11k c000000.wifi: src srng id 150 hp 8190, reap_hp 8190, cur tp 0, cached tp 0 last tp 0 napi processed before 22230ms
[   37.340171] ath11k c000000.wifi: Already processed, so ignoring dma ring caps

WLAN.HK.2.9 latest did enable the radios as normal.
But suddenly rebooted and I did not manage to get that crash log. Sorry about that...

_FailSafe · June 5, 2024, 6:38pm

Here is a comparison between target 5ms + interval 50ms vs. target 8ms + interval 80ms on a GL-MT6000 with WED disabled and AQL low/high both set to 1500:

Target 8ms - Interval 80ms:

Target 5ms - Interval 50ms:

I have some opinions about which seems better, but I'm going to save those thoughts for later. I'm more eager to gather objective feedback from those following this thread.

Spacebar · June 5, 2024, 7:24pm

Hmm... this is interesting. My thoughts when I vote for customize is because qca vs mt. They do things different, and just go for "one fits all" is not something I think would work. There must be a default/fallback too.

And when I've tested different qca WLAN.HK versions I see they differ with bandwidth and latency. I still have issues with AX and iPhone. But that's a different story

And thumbs up for sharing visual data Maybe in the future we can have a service like geekbench for Crusader data. Looking forward to hearing your thoughts.

_FailSafe · June 5, 2024, 7:38pm

Absolutely--without objective comparisons, this is a daunting task due to subjective feedback (and endless Waveform bufferbloat test links that introduce ISP latency into the picture).

Not to take this OT, but...

<OT>
My only real gripe with Crusader right now is that I would desperately love to be able to do one or both of:

Set a consistent Y-axis scale upon the image output. That would make better consistency in comparing one result to another.
Overlay two or more tests on the same screen/image output in a similar way as Flent allows. Again, this would make comparison so much more precise and useful.

</OT>

castiel652 · June 6, 2024, 2:34am

Don't have a ath11k device at the moment but I am tempted to get one.

moeller0 · June 6, 2024, 6:32am

Not that it help much, but since crusader allows to save out the data, there is at least a potential for being able to perform the desired plots outside of crusader. While will require to parse the data first in some other language...

Spacebar · June 8, 2024, 12:48pm

Tested on an IPQ6018, and the same happened once a client connected to the wifi.

kern.alert kernel: [  478.865967] Unable to handle kernel read from unreadable memory at virtual address 0000000000000010
kern.alert kernel: [  478.866026] Mem abort info:
kern.alert kernel: [  478.873854]   ESR = 0x0000000096000005
kern.alert kernel: [  478.876605]   EC = 0x25: DABT (current EL), IL = 32 bits
kern.alert kernel: [  478.880427]   SET = 0, FnV = 0
kern.alert kernel: [  478.885890]   EA = 0, S1PTW = 0
kern.alert kernel: [  478.888761]   FSC = 0x05: level 1 translation fault
kern.alert kernel: [  478.891797] Data abort info:
kern.alert kernel: [  478.896654]   ISV = 0, ISS = 0x00000005, ISS2 = 0x00000000
kern.alert kernel: [  478.899783]   CM = 0, WnR = 0, TnD = 0, TagAccess = 0

dtaht · June 9, 2024, 11:29pm

I am happy to see uptake of crusader... I do not know who the author actually is but i filed a bug on your behalf here: https://github.com/Zoxc/crusader/issues/21

Zoxc · July 30, 2024, 12:51pm

I added an option to increase the Y scale in Crusader. Apparently the graph library I used doesn't support cropping.

I also added the ability to measure latency from an additional separate Crusader instance to the server, which is useful to measure latency impact on a separate Wi-Fi station.

mindwolf · September 7, 2024, 5:41pm

+---------------------------+------------------------------------+----------------------------------+
|        Metric             |   Target 8ms, Interval 80ms        |   Target 5ms, Interval 50ms      |
+---------------------------+------------------------------------+----------------------------------+
| Download Bandwidth        | 500 Mbps (stable)                  | 500 Mbps (stable)                |
| Upload Bandwidth          | 1500+ Mbps (fluctuates)            | 1200 Mbps (more stable)          |
| Latency (Download)        | Low (~20 ms, stable)               | Low (~20 ms, stable)             |
| Latency (Upload)          | Higher peaks (>120 ms)             | Lower peaks (~80 ms)             |
| Total Latency             | High peaks (~120 ms)               | Smoother, fewer high peaks       |
| Packet Loss               | Minor loss during high traffic     | Less loss, better reliability    |
+---------------------------+------------------------------------+----------------------------------+
|      **Pros**             | - Higher max upload bandwidth      | - Lower overall latency          |
|                           | - More relaxed timing (easier to   | - More stable network with less  |
|                           |   handle peak traffic)             |   packet loss                    |
|                           |                                    | - Suitable for latency-sensitive |
|                           |                                    |   applications                   |
+---------------------------+------------------------------------+----------------------------------+
|      **Cons**             | - Higher latency spikes during     | - Slightly lower upload max      |
|                           |   upload-heavy traffic             | - More aggressive timing may     |
|                           | - More variable latency under load |   require fine-tuning            |
+---------------------------+------------------------------------+----------------------------------+