After https://webrtc-review.googlesource.com/c/src/+/329141, best candidate can still be less than acked rate if not_increase_if_inherent_loss_less_than_average_loss, or the selected candidate is 95% of current estimate. This cl/ is ensure the previous cl works as intended. And add unit test.
Bug: webrtc:12707
Change-Id: Ie5683ca8ea51f6d80c4c59cbf08c22e8b24c0cb9
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/329441
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#41298}
These flags were never experimented or launched.
Bug: webrtc:12707
Change-Id: Iefedeade52fdcf7f978894c4bf837261810f41bd
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/329080
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#41265}
To ensure padding, we increase 1 bit instead of 1kbps to avoid that 1kbps adds up over time.
Not have unit test for this, but did manual/hamrit tests many times.
Bug: webrtc:12707
Change-Id: I9b3160ab1808cb3a21ff0609446359a4ec3a4949
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/325520
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#41056}
The change is under field trial use_in_start_phase.
Bug: webrtc:12707
Change-Id: I2ba8245c5d126b3c8a2e54b826853d98aad6e4f9
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/325184
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#41047}
Increasing BWE by 1kbps should be safe/no-op in practice, and it ensures that padding in kIncreasing state will be triggered.
Bug: webrtc:12707
Change-Id: I82493d07a80abd60c93d9cff74baf0a55e77f2b7
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/325286
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#41046}
If we have been sending padding for 1s and estimate still is unchanged, then stop padding by transitioning to decrease state.
Bug: webrtc:12707
Change-Id: I0dca2e5cd98263fc7fae9882c23c21634413c7a0
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/324740
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#41018}
2 main reasons:
1. Packet sizes are much different thus a lost audio packet should not be treated similar to a lost video packet. In low bandwidth/traffic policing scenario, the number of send packet is few, thus the computed loss can be imprecise.
2. Given a candidate bandwidth estimate, the objective function (how good the candidate is) is computed by recomputing loss rate = send rate/estimate bandwith + inherent loss. It means the objective function is byte based rather than packet based.
Potential risk: the current algorithm params are tuned based on packet count, thus it might not work with byte count, which is much higher than packet count.
The change is under field trial and disabled by default.
Bug: webrtc:12707
Change-Id: I8b832e7920d2b4cadcd4a072b3a4d4f26a213a20
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/325065
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#41013}
The initial hold duration is 300ms.
Whenever it enters kDecreasing state, it will double the current hold duration. The hold duration will be reset as soon as the delay based estimate works, e.g. the state is kDelayBased to avoid getting stuck at low bitrate.
Bug: webrtc:12707
Change-Id: I3906ff80b071ba3eb6274b012fb31922f4cbc7b6
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/324304
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40991}
Addes field trial UpperBoundCandidateInAlr to LossBasedBweV2. If an
instant upper bound exist in ALR that are lower than current estimate,
use it as a candidate.
Bug: webrtc:12707
Change-Id: I55595c7225c4289e1bc4edde9d9576e0443d3dce
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/324220
Auto-Submit: Per Kjellander <perkj@webrtc.org>
Reviewed-by: Diep Bui <diepbp@webrtc.org>
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40986}
UsePadding - signals to GoogCC that padding should be used to fill up to
BWE while BWE is ramping up.
Bug: webrtc:12707
Change-Id: I7b4922dff3a83da370c50c567050bfa748190b40
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/324160
Reviewed-by: Diep Bui <diepbp@webrtc.org>
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40979}
Ensure acked bitrate is not used for lower loss based estimate if
estimate improve.
Ensure LossBasedBweV2 is in state DelayBased if reached max rate.
Bug: webrtc:12707
Change-Id: I20230b99e0c2b530570e2f2de8ea88179f795c50
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/324140
Reviewed-by: Diep Bui <diepbp@webrtc.org>
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40977}
Expect OnNetworkAvailabability to be invoked when the transport becomes writable.
Before this change, ProbeController in GoogCC was expected to be created when the transport is writable or explicitly notifed after creation that network is not writable.
Bug: None
Change-Id: I623b1c34e40a82e912f85b92fea49629e7e72d4e
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/323463
Reviewed-by: Diep Bui <diepbp@webrtc.org>
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40975}
The state should be computed from the actual estimate rather than the best estimate candidate. The fix is NOT under field trial.
And some other cleanup:
1. Loss based result will be computed in UpdateBandwidthEstimate method. Currently it is re-computed in GetLossBasedResult.
2. Rename current_estimate to current_best_estimate to avoid misunderstanding that current_estimate is the `final estimate`. The final estimate is computed by applying lower and upper bound on current_best_estimate
3. Remove current_state_. The state is stored directly in loss_based_result_.
Bug: webrtc:12707
Change-Id: Ie612845f907b9e6333fbd8249ddc9b93ad9f8042
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/324022
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40964}
This will be used in an experiment to ramp up BWE when BWE is reduced
due to loss.
Bug: webrtc:12707
Change-Id: I3b78f9dd3fe8ef9f94a9616640ffb8b2225e161e
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/324042
Reviewed-by: Diep Bui <diepbp@webrtc.org>
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40961}
This cl/ makes sure that the estimate cannot go lower than a factor of acked bitrate. The current flag LowerBoundByAckedRateFactor is set to 0, means we dont use it.
Bug: webrtc:12707
Change-Id: I75d5881f0b85a374af3f7039b82c71aee97fb7b0
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/323881
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40958}
This ensure probe results can not be lower than 85% percentage of the
acked bitrate.
Bug: webrtc:11498
Change-Id: I501eeb84f7a049140c45c89e7de7e8080c13f94d
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/324040
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Reviewed-by: Björn Terelius <terelius@webrtc.org>
Auto-Submit: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40957}
MinNumObservations is set to 3 per default as loss based BWE should not be ready if it has few feedbacks. We use a flag, rather than a const since we want to customize it for our unit tests, which often have 1-2 packet feedbacks only, and customize it later in prod if necessary.
Bug: webrtc:12707
Change-Id: Id1cd21aaf6137996de2e51cb5e33fc2a4bb07d8a
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/323780
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40952}
Cleanup field trials for not probing when BWE limited due to high RTT,
loss.
Bug: webrtc:14754, webrtc:12707
Change-Id: Ib664784e321d9284d842ea42a0dd1d8361000f20
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/323640
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Reviewed-by: Diep Bui <diepbp@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40949}
These features are not in use.
Bug: webrtc:12707
Change-Id: Ibe9fcae5e3fd7cb7ca289af80dad8480288c9ba3
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/323601
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Reviewed-by: Diep Bui <diepbp@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40938}
A follow up cl/ is to remove passing upper link capacity from goog_cc to loss_based_bwe_v2.
Bug: webrtc:12707
Change-Id: I45af8ca6e8ba185700d0b7eb57004d2b61edeb9e
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/320780
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40780}
The purpose is to not allow an initial low link capacity estimate to reduce the current estimate.
Only delay overuse detection , low probe results or a loss event can
reduce the estimate.
Bug: webrtc:14392
Change-Id: Ib1618347f2c7681e3bd65d85ee687dec3cd67c97
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/320380
Reviewed-by: Diep Bui <diepbp@google.com>
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40751}
The sequence number is generally not used for the estimation,
but may be used as a tie-breaker when ordering packet feedbacks.
Bug: b/299667054
Change-Id: I52a5145c889c8f6924838667cc267b1cd9565f7b
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/320240
Commit-Queue: Björn Terelius <terelius@webrtc.org>
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40749}
Experiments has not showed significant metric changes. However, simulations has showed that RobustThroughputEstimator better follow the actually receive rate better. Especially during bursts of sent packets. Code is also simpler.
Bug: webrtc:13402 chromium:1411666
Change-Id: I38c309f74e8e1322602196354545b3a465866263
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/318040
Reviewed-by: Björn Terelius <terelius@webrtc.org>
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40653}
Start using RobustThoughputEstimator in DelayBasedBwe test in preparation for making it default.
Experiments has not showed significant metric changes. However, simulations has showed that RobustThroughputEstimator better follow the actually receive rate better. Especially during bursts of sent packets. Code is also simpler.
Bug: webrtc:13402 chromium:1411666
Change-Id: I83cfa1fc15486982b18cc22fbd0752ff59c1c1b4
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/317600
Reviewed-by: Björn Terelius <terelius@webrtc.org>
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#40644}
Ensure probes are not created until after the transport becomes writable even if the network route change.
DTLS negotiation must complete before there is a point in sending probes.
Bug: webrtc:14928
Change-Id: Ib3cb93aef9f38e306b094dd700ed595cf9eb3f32
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/301362
Reviewed-by: Diep Bui <diepbp@webrtc.org>
Commit-Queue: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#39870}
Pass FieldTrialsView by const& to note it can't be null and doesn't need to outlive the constructor
In unittests use AimdRateControl object directly instead of through helpers
Use unit types (TimeDelta, DataRate) directly, reducing their conversion to plain numbers
Replace SimulatedClock with a single Timestamp now variable or constant
Bug: None
Change-Id: I147f85e629b4d8923aa19896ea211a6f9dca1e68
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/299260
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Commit-Queue: Danil Chapovalov <danilchap@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#39707}
Instead of use probe_bitrate as the bandwidth estimate, the change uses probe bitrate as the bandwidth limit.
The experiment is not started, so it does not affect production.
Bug: webrtc:12707
Change-Id: Iefd8cdfe87983057489e551816bf5d4cb38f7c9f
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/296040
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Reviewed-by: Björn Terelius <terelius@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#39603}
Packet pending time should be diffed between max_revc_time and receive time as it is done at line 436. The current implementation makes pending time to be negative.
Bug: webrtc:14850
Change-Id: Ie6590ef11caa67254750591abb6bf72679d76941
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/292461
Commit-Queue: Diep Bui <diepbp@webrtc.org>
Reviewed-by: Per Kjellander <perkj@webrtc.org>
Cr-Commit-Position: refs/heads/main@{#39311}
