Both test and prod setups may use several signaling threads,
this CL prevents race conditions on GenerateUniqueId().
Bug: webrtc:9849
Change-Id: Iaec98b7b4f99729a9ad0642873a5d87de252cb1a
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/147020
Commit-Queue: Yves Gerey <yvesg@google.com>
Commit-Queue: Seth Hampson <shampson@webrtc.org>
Reviewed-by: Seth Hampson <shampson@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28692}
By design:
* OnPacketAdded() is meant to be called on pacer thread.
* Reconfigure() is meant to be called on worker thread.
Thus we guard against race condition on config_ member.
Possible downside: packet filtering based on ssrc might be slowed down.
Bug: webrtc:9849
Change-Id: I734bb9b34b01db160705897adb1b58e866e12639
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146980
Commit-Queue: Yves Gerey <yvesg@google.com>
Reviewed-by: Oskar Sundbom <ossu@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28691}
This is experimental field trial to exclude transport sequence number from FEC packets and should only be used in conjunction with datagram transport. Datagram transport removes transport sequence numbers from RTP packets and uses datagram feedback loop to re-generate RTCP feedback packets, but FEC contorol packets are calculated before sequence number is removed and as a result recovered packets will be corrupt unless we also remove transport sequence number during FEC calculations.
This change is a bit embarrassing, but it was the easiest workaround we found to make FEC work with datagrams. Added TODO to find better long term solution.
TODO(sukhanov): We need to find find better way to implement FEC with datagram transport, probably moving FEC to datagram integration layter. Wealso remove special field trial once we switch datagram path from RTCConfiguration flags to field trial and use the same field trial for FECworkaround.
Bug: webrtc:9719
Change-Id: I1e23c56e3cbaa087460410942fb6c5b4921a763e
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146221
Commit-Queue: Anton Sukhanov <sukhanov@webrtc.org>
Reviewed-by: Steve Anton <steveanton@webrtc.org>
Reviewed-by: Stefan Holmer <stefan@webrtc.org>
Reviewed-by: Bjorn Mellem <mellem@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28686}
These RTP header extensions are used for Unified Plan SDP / BUNDLE and
replace SSRC signaling.
Previously, the RTPSender would attach these header extensions to every
packet when configured. Now, the header extensions will be attached to
every packet until the an RTCP RR is received on that SSRC which
indicates the receiver knows what MID/RID the SSRC is associated with.
This should reduce overhead by 2-4 bytes per packet when the MID header
extension is used and by 4-8 bytes when both header extensions are used.
Bug: webrtc:10078
Change-Id: I5fa3ce28a75224adf11d2792bf4ff8dc76e46d99
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146480
Reviewed-by: Stefan Holmer <stefan@webrtc.org>
Commit-Queue: Steve Anton <steveanton@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28685}
packet_size() includes the size of padding, this means that the size
check might incorrectly not trigger even if the payload is empty. In
turn this means that the ReadBigEndian call might read out of bounds
memory.
Refactored the code to reuse the App parsing code more, eliminating
the risk of this particular kind of error.
Bug: chromium:987507
Change-Id: Id8f3e292c3d30460d3cdb551f0a45070fdf8f022
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146716
Reviewed-by: Stefan Holmer <stefan@webrtc.org>
Commit-Queue: Sebastian Jansson <srte@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28680}
Perf dashboard show a very minor change with the new pacer, for tests
that use flexfec. I have found that previously fec was in fact
prioritized at the same level as video, see eg PacketTypeToPriority()
in RTPSender.
With the new pacer we put fec in between video and padding.
Not sure if this is in fact an actual problem. In the non-loss case
the frame latency should actually be slighly lower, but on the other
hand if we have loss fec won't be applied until after the full frame
has been sent and so we may end up sending NACK before we apply the
FEC and recover a packet.
Just to avoid any problems let's revert to the old behavior.
Bug: webrtc:10633
Change-Id: I9a4210a64165a6e376c0c70ccaa07b0688cc58a5
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146714
Reviewed-by: Sebastian Jansson <srte@webrtc.org>
Commit-Queue: Erik Språng <sprang@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28678}
This reverts commit fab3460a821abe336ab610c6d6dfc0d392dac263.
Reason for revert: fix downstream instead
Original change's description:
> Revert "Reland "Add plumbing of RtpPacketInfos to each AudioFrame as input for SourceTracker.""
>
> This reverts commit 9973933d2e606d64fcdc753acb9ba3afd6e30569.
>
> Reason for revert: breaking downstream projects and not reviewed by direct owners
>
> Original change's description:
> > Reland "Add plumbing of RtpPacketInfos to each AudioFrame as input for SourceTracker."
> >
> > This reverts commit 24192c267a40eb7d6b1850489ccdbf7a84f8ff0f.
> >
> > Reason for revert: Analyzed the performance regression in more detail.
> >
> > Most of the regression comes from the extra RtpPacketInfos-related memory allocations in every `NetEq::GetAudio()` call. Commit 1796a820f60cb9429bf4bcf13a40a41794ac8fb0 has removed roughly 2/3rds of the extra allocations from the impacted perf tests. Remaining perf impact is expected to be about "8 microseconds of CPU time per second" on the Linux benchmarking machines and "15 us per second" on Windows/Mac.
> >
> > There are options to optimize further but they are unlikely worth doing. Note for example that `NetEqPerformanceTest` uses the PCM codec while the real-world use cases would likely use the much heavier Opus codec. The numbers from `OpusSpeedTest` and `NetEqPerformanceTest` suggest that Opus decoding is about 10x as expensive as NetEq overall.
> >
> > Original change's description:
> > > Revert "Add plumbing of RtpPacketInfos to each AudioFrame as input for SourceTracker."
> > >
> > > This reverts commit 3e8ef940fe86cf6285afb80e68d2a0bedc631b9f.
> > >
> > > Reason for revert: This CL causes a performance regression in NetEq, see https://bugs.chromium.org/p/chromium/issues/detail?id=982260.
> > >
> > > Original change's description:
> > > > Add plumbing of RtpPacketInfos to each AudioFrame as input for SourceTracker.
> > > >
> > > > This change adds the plumbing of RtpPacketInfo from ChannelReceive::OnRtpPacket() to ChannelReceive::GetAudioFrameWithInfo() for audio. It is a step towards replacing the non-spec compliant ContributingSources that updates itself at packet-receive time, with the spec-compliant SourceTracker that will update itself at frame-delivery-to-track time.
> > > >
> > > > Bug: webrtc:10668
> > > > Change-Id: I03385d6865bbc7bfbef7634f88de820a934f787a
> > > > Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/139890
> > > > Reviewed-by: Stefan Holmer <stefan@webrtc.org>
> > > > Reviewed-by: Minyue Li <minyue@webrtc.org>
> > > > Commit-Queue: Chen Xing <chxg@google.com>
> > > > Cr-Commit-Position: refs/heads/master@{#28434}
> > >
> > > TBR=kwiberg@webrtc.org,stefan@webrtc.org,minyue@webrtc.org,chxg@google.com
> > >
> > > Bug: webrtc:10668, chromium:982260
> > > Change-Id: I5e2cfde78c59d1123e21869564d76ed3f6193a5c
> > > Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/145339
> > > Reviewed-by: Ivo Creusen <ivoc@webrtc.org>
> > > Commit-Queue: Ivo Creusen <ivoc@webrtc.org>
> > > Cr-Commit-Position: refs/heads/master@{#28561}
> >
> > TBR=kwiberg@webrtc.org,stefan@webrtc.org,ivoc@webrtc.org,minyue@webrtc.org,chxg@google.com
> >
> > # Not skipping CQ checks because original CL landed > 1 day ago.
> >
> > Bug: webrtc:10668, chromium:982260
> > Change-Id: Ie375a0b327ee368317bf3a04b2f1415c3a974470
> > Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146707
> > Reviewed-by: Stefan Holmer <stefan@webrtc.org>
> > Commit-Queue: Chen Xing <chxg@google.com>
> > Cr-Commit-Position: refs/heads/master@{#28664}
>
> TBR=kwiberg@webrtc.org,stefan@webrtc.org,ivoc@webrtc.org,minyue@webrtc.org,chxg@google.com
>
> Change-Id: I652cb0814d83b514d3bee34e65ca3bb693099b22
> No-Presubmit: true
> No-Tree-Checks: true
> No-Try: true
> Bug: webrtc:10668, chromium:982260
> Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146712
> Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
> Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
> Cr-Commit-Position: refs/heads/master@{#28671}
TBR=alessiob@webrtc.org,kwiberg@webrtc.org,stefan@webrtc.org,ivoc@webrtc.org,minyue@webrtc.org,chxg@google.com
Change-Id: Id43b7b3da79b4f48004b41767482bae1c1fa1e16
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: webrtc:10668, chromium:982260
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146713
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28672}
This reverts commit 9973933d2e606d64fcdc753acb9ba3afd6e30569.
Reason for revert: breaking downstream projects and not reviewed by direct owners
Original change's description:
> Reland "Add plumbing of RtpPacketInfos to each AudioFrame as input for SourceTracker."
>
> This reverts commit 24192c267a40eb7d6b1850489ccdbf7a84f8ff0f.
>
> Reason for revert: Analyzed the performance regression in more detail.
>
> Most of the regression comes from the extra RtpPacketInfos-related memory allocations in every `NetEq::GetAudio()` call. Commit 1796a820f60cb9429bf4bcf13a40a41794ac8fb0 has removed roughly 2/3rds of the extra allocations from the impacted perf tests. Remaining perf impact is expected to be about "8 microseconds of CPU time per second" on the Linux benchmarking machines and "15 us per second" on Windows/Mac.
>
> There are options to optimize further but they are unlikely worth doing. Note for example that `NetEqPerformanceTest` uses the PCM codec while the real-world use cases would likely use the much heavier Opus codec. The numbers from `OpusSpeedTest` and `NetEqPerformanceTest` suggest that Opus decoding is about 10x as expensive as NetEq overall.
>
> Original change's description:
> > Revert "Add plumbing of RtpPacketInfos to each AudioFrame as input for SourceTracker."
> >
> > This reverts commit 3e8ef940fe86cf6285afb80e68d2a0bedc631b9f.
> >
> > Reason for revert: This CL causes a performance regression in NetEq, see https://bugs.chromium.org/p/chromium/issues/detail?id=982260.
> >
> > Original change's description:
> > > Add plumbing of RtpPacketInfos to each AudioFrame as input for SourceTracker.
> > >
> > > This change adds the plumbing of RtpPacketInfo from ChannelReceive::OnRtpPacket() to ChannelReceive::GetAudioFrameWithInfo() for audio. It is a step towards replacing the non-spec compliant ContributingSources that updates itself at packet-receive time, with the spec-compliant SourceTracker that will update itself at frame-delivery-to-track time.
> > >
> > > Bug: webrtc:10668
> > > Change-Id: I03385d6865bbc7bfbef7634f88de820a934f787a
> > > Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/139890
> > > Reviewed-by: Stefan Holmer <stefan@webrtc.org>
> > > Reviewed-by: Minyue Li <minyue@webrtc.org>
> > > Commit-Queue: Chen Xing <chxg@google.com>
> > > Cr-Commit-Position: refs/heads/master@{#28434}
> >
> > TBR=kwiberg@webrtc.org,stefan@webrtc.org,minyue@webrtc.org,chxg@google.com
> >
> > Bug: webrtc:10668, chromium:982260
> > Change-Id: I5e2cfde78c59d1123e21869564d76ed3f6193a5c
> > Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/145339
> > Reviewed-by: Ivo Creusen <ivoc@webrtc.org>
> > Commit-Queue: Ivo Creusen <ivoc@webrtc.org>
> > Cr-Commit-Position: refs/heads/master@{#28561}
>
> TBR=kwiberg@webrtc.org,stefan@webrtc.org,ivoc@webrtc.org,minyue@webrtc.org,chxg@google.com
>
> # Not skipping CQ checks because original CL landed > 1 day ago.
>
> Bug: webrtc:10668, chromium:982260
> Change-Id: Ie375a0b327ee368317bf3a04b2f1415c3a974470
> Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146707
> Reviewed-by: Stefan Holmer <stefan@webrtc.org>
> Commit-Queue: Chen Xing <chxg@google.com>
> Cr-Commit-Position: refs/heads/master@{#28664}
TBR=kwiberg@webrtc.org,stefan@webrtc.org,ivoc@webrtc.org,minyue@webrtc.org,chxg@google.com
Change-Id: I652cb0814d83b514d3bee34e65ca3bb693099b22
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Bug: webrtc:10668, chromium:982260
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146712
Reviewed-by: Alessio Bazzica <alessiob@webrtc.org>
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28671}
In some situation, we disable simulcast in the encoder pipeline without
changing the sender's RIDs and it would crash.
This should only happen now when requesting simulcast with VP9 codec,
for which you currently get SVC instead.
Bug: webrtc:10660
Change-Id: I4f3b3d7760aded8f0769f8357c03ed8580ea46fa
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/145336
Reviewed-by: Sebastian Jansson <srte@webrtc.org>
Commit-Queue: Florent Castelli <orphis@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28669}
- Make rtp_analyzer work with a single SSRC
- Simplify rtp_analyzer.sh (it allows to run the python script
from any directory)
- Update README.md (simplified, added missing dependency)
Bug: webrtc:10829
Change-Id: Idb82e7228918a973778762a39b732ce3b26b6bbc
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146711
Reviewed-by: Artem Titov <titovartem@webrtc.org>
Reviewed-by: Mirko Bonadei <mbonadei@webrtc.org>
Commit-Queue: Alessio Bazzica <alessiob@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28668}
In https://webrtc-review.googlesource.com/c/src/+/138275
the congestion window was recalculated during OnProcessInterval, as
to consider the case when downlink is down. However, this update
was not propagated to the congestion window pusback controller,
nor returned in the update.
This patch fixes that issue, as well as adding two tests to ensure
the behaviour works as expected.
Bug: None
Change-Id: Ic126d929dc7a7a3393a2f34a4682eea1ee1f2240
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146704
Commit-Queue: Evan Shrubsole <eshr@google.com>
Reviewed-by: Sebastian Jansson <srte@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28667}
Some applications are having issues controlling how many active simulcast
layers they have and configuring them when the source track changes size
and rely on the old behavior automatically reducing the layer count.
Using the WebRTC-LegacySimulcastLayerLimit field trial, they can get back
the old behavior until they transition to the newer API.
Bug: webrtc:8785
Change-Id: I92d4dcd62b79a483a6a8867f97c5f502c6aa4db7
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146709
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Florent Castelli <orphis@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28666}
This reverts commit 24192c267a40eb7d6b1850489ccdbf7a84f8ff0f.
Reason for revert: Analyzed the performance regression in more detail.
Most of the regression comes from the extra RtpPacketInfos-related memory allocations in every `NetEq::GetAudio()` call. Commit 1796a820f60cb9429bf4bcf13a40a41794ac8fb0 has removed roughly 2/3rds of the extra allocations from the impacted perf tests. Remaining perf impact is expected to be about "8 microseconds of CPU time per second" on the Linux benchmarking machines and "15 us per second" on Windows/Mac.
There are options to optimize further but they are unlikely worth doing. Note for example that `NetEqPerformanceTest` uses the PCM codec while the real-world use cases would likely use the much heavier Opus codec. The numbers from `OpusSpeedTest` and `NetEqPerformanceTest` suggest that Opus decoding is about 10x as expensive as NetEq overall.
Original change's description:
> Revert "Add plumbing of RtpPacketInfos to each AudioFrame as input for SourceTracker."
>
> This reverts commit 3e8ef940fe86cf6285afb80e68d2a0bedc631b9f.
>
> Reason for revert: This CL causes a performance regression in NetEq, see https://bugs.chromium.org/p/chromium/issues/detail?id=982260.
>
> Original change's description:
> > Add plumbing of RtpPacketInfos to each AudioFrame as input for SourceTracker.
> >
> > This change adds the plumbing of RtpPacketInfo from ChannelReceive::OnRtpPacket() to ChannelReceive::GetAudioFrameWithInfo() for audio. It is a step towards replacing the non-spec compliant ContributingSources that updates itself at packet-receive time, with the spec-compliant SourceTracker that will update itself at frame-delivery-to-track time.
> >
> > Bug: webrtc:10668
> > Change-Id: I03385d6865bbc7bfbef7634f88de820a934f787a
> > Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/139890
> > Reviewed-by: Stefan Holmer <stefan@webrtc.org>
> > Reviewed-by: Minyue Li <minyue@webrtc.org>
> > Commit-Queue: Chen Xing <chxg@google.com>
> > Cr-Commit-Position: refs/heads/master@{#28434}
>
> TBR=kwiberg@webrtc.org,stefan@webrtc.org,minyue@webrtc.org,chxg@google.com
>
> Bug: webrtc:10668, chromium:982260
> Change-Id: I5e2cfde78c59d1123e21869564d76ed3f6193a5c
> Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/145339
> Reviewed-by: Ivo Creusen <ivoc@webrtc.org>
> Commit-Queue: Ivo Creusen <ivoc@webrtc.org>
> Cr-Commit-Position: refs/heads/master@{#28561}
TBR=kwiberg@webrtc.org,stefan@webrtc.org,ivoc@webrtc.org,minyue@webrtc.org,chxg@google.com
# Not skipping CQ checks because original CL landed > 1 day ago.
Bug: webrtc:10668, chromium:982260
Change-Id: Ie375a0b327ee368317bf3a04b2f1415c3a974470
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146707
Reviewed-by: Stefan Holmer <stefan@webrtc.org>
Commit-Queue: Chen Xing <chxg@google.com>
Cr-Commit-Position: refs/heads/master@{#28664}
This makes the code path where packets are directly owned by PacedSender
rather that being temporarily put in the RtpPacketHistory the default.
Functionally, this should essentially be a noop, with only minor timing
differences.
The old code-path will stay around for a short while and then be
removed once we are certain there are no regressions.
Bug: webrtc:10633
Change-Id: Id6360dea48fd0c9d46fde6f5eee93726d4f11d13
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146212
Commit-Queue: Erik Språng <sprang@webrtc.org>
Reviewed-by: Stefan Holmer <stefan@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28660}
If the SSRC of an RTP module is changed at runtime, we may get conflicts
with packets already there. Eg:
* Put seq# 123 in the history for SSRC 1.
* Change the SSRC to 2.
* Send a NACK for seq# 123 from SSRC 2.
Currently, we will respond with the packet belonging to SSRC 1 (and not
if the NACK specifies SSRC 1, to boot).
We can gen similar issues if the sequence number is changed, where
half frame are left in the buffer.
In these cases, the stream is likely being reset so we should just
clear the packet history too.
Bug: webrtc:10794
Change-Id: I28147c2532cf1c78840d4808c4366d4a647541f7
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/145729
Commit-Queue: Erik Språng <sprang@webrtc.org>
Reviewed-by: Stefan Holmer <stefan@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28658}
This CL fixes two issues related to the TransmissionOffset header
extension and the new (not yet active) pacer mode.
Previously capture time (if unset) would be populated when put into the
packet history before entering the pacer. Since the pacer now owns the
packets, this does not occur until packet is actually sent, if at all.
Capture has really nothing to do with the packet history, this should
be set by the RtpSender pre-pacing instead.
Furthermore, for retransmissions the old path would take the capture
time from the original packet, build the RTX-wrapped retransmission and
set the toffset extension of the RTX packet using that captured capture
time. Since RTX packets are now fully built before the pacer, this does
not work, and we need to transfer the capture time from the original to
the RTX packet instead.
Bug: webrtc:10633
Change-Id: I031e8b6cc4ab20fb094dbd46720829b78951e7f9
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146218
Commit-Queue: Erik Språng <sprang@webrtc.org>
Reviewed-by: Stefan Holmer <stefan@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28657}
This is intended to by used for visualizing catagorical data, i.e. mapping
numerical enum values to string labels.
Bug: webrtc:10623
Change-Id: Ic9c3da9a3874f479c07412f394a774ae90fd3d7e
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/145408
Reviewed-by: Sebastian Jansson <srte@webrtc.org>
Commit-Queue: Björn Terelius <terelius@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28656}
This adds the RemoteEstimate rtcp packet and wires it up to GoogCC where
it's used to improve congestion controller behavior.
The functionality is negotiated using SDP.
It's added with a field trial that allow disabling the functionality in
case there's any issues.
Bug: webrtc:10742
Change-Id: I1ea8e4216a27cd2b00505c99b42d1e38726256c8
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/146602
Reviewed-by: Stefan Holmer <stefan@webrtc.org>
Reviewed-by: Erik Språng <sprang@webrtc.org>
Commit-Queue: Sebastian Jansson <srte@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#28654}
