[feat/gcc] Implement Bandwidth Estimation #71

mengelbart · 2021-11-02T10:11:38Z

Description

This pull request tracks the implementation progress for bandwidth estimation using Google Congestion Control. It will introduce interfaces that can hopefully be reused by other bandwidth estimators in the future.

Reference issue

Implements the GCC related tasks of #25

TODO:

codecov · 2021-12-06T15:59:51Z

Codecov Report

Merging #71 (c1f02ed) into master (ec82d53) will decrease coverage by 11.26%.
The diff coverage is 49.08%.

❗ Current head c1f02ed differs from pull request most recent head 04c55ce. Consider uploading reports for the commit 04c55ce to get more accurate results

@@             Coverage Diff             @@
##           master      #71       +/-   ##
===========================================
- Coverage   79.38%   68.12%   -11.27%     
===========================================
  Files          32       51       +19     
  Lines        1528     2453      +925     
===========================================
+ Hits         1213     1671      +458     
- Misses        244      699      +455     
- Partials       71       83       +12

Flag	Coverage Δ
go	`68.12% <49.08%> (-11.27%)`	⬇️
wasm	`66.89% <49.08%> (-10.79%)`	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
pkg/gcc/acknowledgment.go	`0.00% <0.00%> (ø)`
pkg/gcc/leaky_bucket_pacer.go	`0.00% <0.00%> (ø)`
pkg/gcc/loss_based_bwe.go	`0.00% <0.00%> (ø)`
pkg/gcc/noop_pacer.go	`0.00% <0.00%> (ø)`
pkg/gcc/send_side_bwe.go	`0.00% <0.00%> (ø)`
pkg/gcc/usage.go	`0.00% <0.00%> (ø)`
pkg/gcc/delay_based_bwe.go	`12.12% <12.12%> (ø)`
pkg/gcc/state.go	`16.21% <16.21%> (ø)`
pkg/gcc/arrival_group.go	`41.66% <41.66%> (ø)`
pkg/gcc/minmax.go	`57.14% <57.14%> (ø)`
... and 10 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ec82d53...04c55ce. Read the comment docs.

pkg/cc/interceptor.go

pkg/gcc/adaptive_threshold.go

pkg/gcc/rate_controller.go

pkg/gcc/state.go

pkg/gcc/usage.go

pkg/gcc/static_threshold.go

pkg/gcc/send_side_bwe.go

pkg/gcc/loss_based_bwe.go

Sean-Der · 2022-01-14T03:26:46Z

pkg/gcc/acknowledgment.go

@@ -0,0 +1,26 @@
+package gcc


Do you think we could move this to internal/cc

This seems like something we could share between congestion controllers. Would love to see this directory be dedicated to gcc specific stuff!

Yes, I think we could move it out of this package, but I'm not sure what the right place would be, internal/cc would mean that pkg/gcc can basically only be used from this module, right? I think it would be nice to allow using pkg/gcc without interceptors?

Oh yes! pkg/cc maybe?

Yes, I think that would work, but currently I use a default BandwidthEstimator in the interceptor, so that would create an import cycle. But maybe there's a better way to set the default estimator without including the package just for that?

*and the default uses gcc.NewSendSideBWE()

We could maybe do twcc for now and fix in the future?

Yes, sounds good!

Implements a first version of Google Congestion Control as described in https://datatracker.ietf.org/doc/html/draft-ietf-rmcat-gcc-02

alexpokotilo · 2024-01-24T10:47:21Z

pkg/gcc/kalman.go

+	zms := float64(z.Microseconds()) / 1000.0
+
+	if !k.disableMeasurementUncertaintyUpdates {
+		alpha := math.Pow((1 - chi), 30.0/(1000.0*5*float64(time.Millisecond)))


@mengelbart
I was not able to get why we use "5*float64(time.Millisecond)"
for f_max in
alpha = (1-chi)^(30/(1000 * f_max)) calculation.
This is important part and has impact on uncertainty calculation. Please help to clue

alexpokotilo · 2024-01-24T10:50:05Z

pkg/gcc/adaptive_threshold.go

+	if a.numDeltas < 2 {
+		return usageNormal, estimate, a.max
+	}
+	t := time.Duration(minInt(a.numDeltas, maxDeltas)) * estimate


@mengelbart
why do we multiply estimate(m_hat(i)) from Kalman filter by up to 60 ?
I was not able to find this in standard and/or https://c3lab.poliba.it/images/6/65/Gcc-analysis.pdf

mengelbart force-pushed the feat/gcc branch from 6793c6c to 3767e28 Compare November 5, 2021 09:52

mengelbart force-pushed the feat/gcc branch 4 times, most recently from fb64001 to 221d52b Compare November 19, 2021 17:10

mengelbart force-pushed the feat/gcc branch from 066dc50 to e43d585 Compare November 23, 2021 20:50

pionbot force-pushed the feat/gcc branch from e43d585 to f0a0339 Compare November 23, 2021 20:50

mengelbart force-pushed the feat/gcc branch from f0a0339 to e99a4ca Compare November 23, 2021 21:43

pionbot force-pushed the feat/gcc branch from e99a4ca to 33e3463 Compare November 23, 2021 21:43

mengelbart force-pushed the feat/gcc branch from 33e3463 to 39ece9d Compare November 23, 2021 21:44

mengelbart force-pushed the feat/gcc branch from 39ece9d to 19e6dfa Compare December 6, 2021 15:58

mengelbart force-pushed the feat/gcc branch 3 times, most recently from d925ea0 to 846af2f Compare December 10, 2021 16:48

mengelbart force-pushed the feat/gcc branch from 95c98c7 to f271b1a Compare December 16, 2021 12:05

mengelbart force-pushed the feat/gcc branch 6 times, most recently from 14eb822 to dce3ab5 Compare January 13, 2022 19:10

mengelbart requested a review from Sean-Der January 13, 2022 19:22

mengelbart marked this pull request as ready for review January 13, 2022 19:22