In order to not make this CL too large I have broken it down into at least two steps. In this CL we only propagate the pacing information part of the way:
webrtc::PacedSender::Process <--- propagate from here
webrtc::PacedSender::SendPacket
webrtc::PacketRouter::TimeToSendPacket
webrtc::ModuleRtpRtcpImpl::TimeToSendPacket <--- to here
webrtc::RTPSender::TimeToSendPacket
webrtc::RTPSender::PrepareAndSendPacket
webrtc::RTPSender::AddPacketToTransportFeedback
webrtc::TransportFeedbackAdapter::AddPacket
webrtc::SendTimeHistory::AddAndRemoveOld <--- goal is to propagte it here
BUG=webrtc:6822
Review-Url: https://codereview.webrtc.org/2628563003
Cr-Commit-Position: refs/heads/master@{#16664}
Previously BirateProber was calculating delay between probes based on
the size of the previous probe. Because of that the actual sent bitrate
can deviate greatly from the target value. With this change it uses
total number of bytes in the cluster to estimate delay before each
probe.
BUG=webrtc:6952
Review-Url: https://codereview.webrtc.org/2613543003
Cr-Original-Commit-Position: refs/heads/master@{#15971}
Committed: 599c5011e7
Review-Url: https://codereview.webrtc.org/2613543003
Cr-Commit-Position: refs/heads/master@{#16127}
Reason for revert:
Speculative revert.
Linux memcheck bot started failing a lot at the time of this cl. Doesn't look related at first glance, but we don't have another lead yet.
Original issue's description:
> Fix BitrateProber to match the requested bitrate more precisely
>
> Previously BirateProber was calculating delay between probes based on
> the size of the previous probe. Because of that the actual sent bitrate
> can deviate greatly from the target value. With this change it uses
> total number of bytes in the cluster to estimate delay before each
> probe.
>
> BUG=webrtc:6952
>
> Review-Url: https://codereview.webrtc.org/2613543003
> Cr-Commit-Position: refs/heads/master@{#15971}
> Committed: 599c5011e7TBR=philipel@webrtc.org,stefan@webrtc.org,sergeyu@chromium.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=webrtc:6952
Review-Url: https://codereview.webrtc.org/2626473004
Cr-Commit-Position: refs/heads/master@{#15979}
Previously BirateProber was calculating delay between probes based on
the size of the previous probe. Because of that the actual sent bitrate
can deviate greatly from the target value. With this change it uses
total number of bytes in the cluster to estimate delay before each
probe.
BUG=webrtc:6952
Review-Url: https://codereview.webrtc.org/2613543003
Cr-Commit-Position: refs/heads/master@{#15971}
Instead of having to specify a bitrate and how many packets to use,
the BitrateProber will now use the bitrate to calculate how many
bytes it will use to probe that bitrate instead.
For now, |kMinProbeDurationMs| is set to 15 ms which means that probing
at 1900 kbps will result in 1900/8*0.015 = 3.5 kB used. Since we can
expect packets to be smaller at the beginning of a stream (500 to 700
bytes) this will result in 7 to 5 packets sent for that bitrate, and
should work very similar to how the current initial probing works.
A minimum of 5 packets are always sent.
BUG=webrtc:6822
Review-Url: https://codereview.webrtc.org/2609113003
Cr-Commit-Position: refs/heads/master@{#15899}
Currently, BitrateProber does not scale higher than 2 Mbps to 6 Mbps. The actual
number is dependent on the size of the last packet. If a packet of around 250
bytes is used for probing, it fails above 2 Mbps.
BitrateProber now provides a recommendation on probe size instead of a
packet size. PacedSender utilizes this to decide on the number of packets
per probe. This enables BitrateProber to scale up-to higher bitrates.
Tests with chromoting show it stalls at about 10 Mbps (perhaps due to the
limitation on the simulation pipeline to deliver packets).
BUG=webrtc:6332
Review-Url: https://codereview.webrtc.org/2347023002
Cr-Commit-Position: refs/heads/master@{#14503}
The intended behavior of the pacer is that it should send audio (i.e high priority) packets immediatelly even if the pacer is paused and even if the available media budget has been used. The existing code will not send audio if the pacing budget has been used. It will normally send audio when the pacer is paused, but if the media budget was already used up when the pacer was paused, it would neither send audio nor update the budget.
Updated the implementatation and the unit test. Added logging for when the pacer is paused, and when it receives a zero bitrate estimate.
BUG=webrtc:6155
Review-Url: https://codereview.webrtc.org/2193183003
Cr-Commit-Position: refs/heads/master@{#13592}
Retransmissions are supposed to be sent before normal packets by the pacer, but the current implementation will only use it if the second packet is a retransmission and the first packet is not. It misses the case where the first packet is retransmission and the second packet is not.
This CL fixes the comparator and adds a unit test.
Also changed the SendAndExpectPacket function to propagate the retransmission flag to the expectations. Previously, all packets were expected to be normal packets.
BUG=webrtc:6124
Review-Url: https://codereview.webrtc.org/2156063004
Cr-Commit-Position: refs/heads/master@{#13502}
1. It moves calculation of the needed padding to VideoSendStream instead of ViEEncoder and only does it once per send Stream instead of every time the network estimate changes.
2. The maximum amount of padding sent was prior to this cl calculated and updated based on network estimate changes. However, it can only change based on encoder configuration changes and if send streams are added or removed. This cl change the VideoSendStream/VieEncoder to notify the BitrateAllocator of changes to the needed padding bitrate and for BitrateAllocator to notify Call of these changes.
3. Fixed an issue in the SendPacer where it could send a padding packet before sending a real packet. This caused the test EndToEndTest.RestartingSendStreamPreservesRtpStatesWithRtx to fail with these refactorings since the pacer suddenly could send a padding packet before the encoder had produced its first frame.
BUG=webrtc:5687
Review-Url: https://codereview.webrtc.org/1993113003
Cr-Commit-Position: refs/heads/master@{#13149}
This reverts commit e30c27205148b34ba421184efe65f6a0780b436d (https://codereview.webrtc.org/1958053002/)
Original reverted cl is in patch set #1.
Changes in following patch sets.
The cl now also make sure SendPacer starts with the configured bitrate provided in a call to CongestionController::SetBweBitrates)()
It turns out that the failing tests in 609816 is due to a bug in the current code that runs the proper at 300kbit regardless of configured start bitrate.
Original cl description:
Remove SendPacer from ViEEncoder
This CL moves the logic where the ViEEncoder pause if the pacer is full to the BitrateController. If the queue is full, the controller reports a bitrate of zero to Call (and BitrateAllocator)
BUG=chromium:609816, webrtc:5687
TBR=mflodman@webrtc.org
NOTRY=True // Due to bug in android_x86 cq builder....
Review-Url: https://codereview.webrtc.org/1958113003
Cr-Commit-Position: refs/heads/master@{#12688}
This reverts commit 825eb58d59940a4c3c9837595c4b3b07059c93ca.
This Relands the cl reviewed in https://codereview.webrtc.org/1917793002/
patchset #1 is a pure reland.
patchset #2 fix an overflow in BitrateProber that caused WebRtcVideoChannel2BaseTest.TwoStreamsSendAndReceive to fail.
Original cl description:
Remove SendPacer from ViEEncoder
This CL moves the logic where the ViEEncoder pause if the pacer is full to the BitrateController. If the queue is full, the controller reports a bitrate of zero to Call (and BitrateAllocator)
R=stefan@webrtc.orgTBR=mflodman@webrtc.org
BUG=webrtc:5687
Review URL: https://codereview.webrtc.org/1947873002 .
Cr-Commit-Position: refs/heads/master@{#12630}
This CL moves the logic where the ViEEncoder pause if the pacer is full to the BitrateController. If the queue is full, the controller reports a bitrate of zero to Call (and BitrateAllocator)
BUG=webrtc:5687
Review-Url: https://codereview.webrtc.org/1917793002
Cr-Commit-Position: refs/heads/master@{#12620}
Except in places where this would break out-of-tree code,
such as Chromium.
BUG=webrtc:5520
Review URL: https://codereview.webrtc.org/1785173002
Cr-Commit-Position: refs/heads/master@{#12037}
Reason for revert:
Revert breaks other uses, a fix will be rolled into Chromium instead.
Original issue's description:
> Revert of Remove ignored return code from modules. (patchset #3 id:40001 of https://codereview.webrtc.org/1703833002/ )
>
> Reason for revert:
> Breaks Chromium.
>
> Original issue's description:
> > Remove ignored return code from modules.
> >
> > ModuleProcessImpl doesn't act on return codes and having them around is
> > confusing (it's unclear what an error return code here would do even).
> >
> > BUG=
> > R=tommi@webrtc.org
> >
> > Committed: f14c47a58c
>
> TBR=tommi@webrtc.org,pbos@webrtc.org
> # Skipping CQ checks because original CL landed less than 1 days ago.
> NOPRESUBMIT=true
> NOTREECHECKS=true
> NOTRY=true
> BUG=
>
> Committed: https://crrev.com/da33a8a2a22f6d19ba2a8cce963beafbdbaa8fd8
> Cr-Commit-Position: refs/heads/master@{#11761}
TBR=tommi@webrtc.org,torbjorng@webrtc.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=
Review URL: https://codereview.webrtc.org/1737013002
Cr-Commit-Position: refs/heads/master@{#11762}
Reason for revert:
Breaks Chromium.
Original issue's description:
> Remove ignored return code from modules.
>
> ModuleProcessImpl doesn't act on return codes and having them around is
> confusing (it's unclear what an error return code here would do even).
>
> BUG=
> R=tommi@webrtc.org
>
> Committed: f14c47a58cTBR=tommi@webrtc.org,pbos@webrtc.org
# Skipping CQ checks because original CL landed less than 1 days ago.
NOPRESUBMIT=true
NOTREECHECKS=true
NOTRY=true
BUG=
Review URL: https://codereview.webrtc.org/1736663004
Cr-Commit-Position: refs/heads/master@{#11761}
ModuleProcessImpl doesn't act on return codes and having them around is
confusing (it's unclear what an error return code here would do even).
BUG=
R=tommi@webrtc.org
Review URL: https://codereview.webrtc.org/1703833002 .
Cr-Commit-Position: refs/heads/master@{#11747}
We should only account for audio packets in the pacer budget if we also
are allocating bandwidth for the audio streams.
BUG=chromium:567659,webrtc:5263
R=mflodman@webrtc.org
Review URL: https://codereview.webrtc.org/1524763002 .
Cr-Commit-Position: refs/heads/master@{#11053}
Currently we limit the enocder so that frames aren't encoded if the
expected pacer queue is longer than 2s. However, if the queue is full
and the bitrate suddenly drops (or there is a large overshoot), the
queue time can be long than the limit.
This CL allows the pacer to temporarily boost the pacing bitrate over
the 2s window.
BUG=
Review URL: https://codereview.webrtc.org/1412293003
Cr-Commit-Position: refs/heads/master@{#10729}
Since the pacer is always enabled, removing enable/disable which makes
all packet queueing succeed. Also renaming one of the ::SendPackets
::InsertPacket to avoid confusion.
BUG=webrtc:1695, webrtc:2629
R=stefan@webrtc.org
Review URL: https://codereview.webrtc.org/1392513002 .
Cr-Commit-Position: refs/heads/master@{#10211}
To make this possible padding only packets will have the same timestamp
as the previously sent media packet, as long as RTX is not enabled. This
has the side effect that if we send only padding for a long time without
sending media, a receive-side jitter buffer could potentially overflow.
In practice this shouldn't be an issue, partly because RTX is recommended and
used by default, but also because padding typically is terminated before being
received by a client. It is also not an issue for bandwidth estimation as long
as abs-send-time is used instead of toffset.
BUG=chromium:425925
R=mflodman@webrtc.org, sprang@webrtc.org, tommi@webrtc.org
Review URL: https://codereview.webrtc.org/1327933003 .
Cr-Commit-Position: refs/heads/master@{#9984}
Packet queue in the paced sender is now based on a priority queue rather than having a separate fifo-queue per priority level. This allows more flexible sorting and cleaner usage.
Packets with earlier capture times are now prioritized higher. In situations with high packet loss, the queue might contain packets from several subsequent frames. Retransmit packets from the earlier frames first, since the later ones will probably be dependent on these.
Also, don't force sending of packets after a certain time of inactivity or when packets grow too old, since this was causing consistent overuse on poor connections. Instead, drop frames in vie encoder if pacer queue is too long.
BUG=
R=mflodman@webrtc.org, stefan@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/27869004
git-svn-id: http://webrtc.googlecode.com/svn/trunk@7617 4adac7df-926f-26a2-2b94-8c16560cd09d
