Add experimental 'backlog' heuristic for measuring encoder utilization.

Add experimental 'backlog' heuristic for measuring encoder utilization.

The existing 'deadline' heuristic has proven to be very flawed on some
systems, for reasons explained in the bug. This change adds a new
experimental heuristic, activated via a new command-line switch, for lab
testing: --cast-encoder-util-heuristic=backlog<N>

If the new heuristic proves to work well across our lab's gamut of test
systems, a future change will remove the command-line switch and make
the new heuristic the default.

BUG=chrome-os-partner:54806
Committed: https://crrev.com/70c68b6a6eb9a3c46662b5ed112f36d53f8e9813
Cr-Commit-Position: refs/heads/master@{#403378}

[miu, 2016-06-29 22:57:45]

Description was changed from

==========
Add experimental 'backlog' heuristic for measuring encoder utilization.

The existing 'deadline' heuristic has proven to be very flawed on some
systems, for reasons explained in the bug. This change adds a new
experimental heuristic, activated via a new command-line switch, for lab
testing: --cast-encoder-util-heuristic=backlog<N>

If the new heuristic proves to work well across our lab's gamut of test
systems, a future change will remove the command-line switch and make
the new heuristic the default.

BUG=https://code.google.com/p/chrome-os-partner/issues/detail?id=54806
==========

to

==========
Add experimental 'backlog' heuristic for measuring encoder utilization.

The existing 'deadline' heuristic has proven to be very flawed on some
systems, for reasons explained in the bug. This change adds a new
experimental heuristic, activated via a new command-line switch, for lab
testing: --cast-encoder-util-heuristic=backlog<N>

If the new heuristic proves to work well across our lab's gamut of test
systems, a future change will remove the command-line switch and make
the new heuristic the default.

BUG=chrome-os-partner:54806
==========

[miu, 2016-06-29 22:57:45]

miu@chromium.org changed reviewers:
+ chcunningham@chromium.org, sievers@chromium.org, xjz@chromium.org

[miu, 2016-06-29 23:00:06]

chcunningham: Need OWNERS RS for adding a switch to media/base/media_switches.cc

sievers: Need OWNERS RS for content/.../render_process_host_impl.cc (trivial
addition to propagate command-line flag to render process).

xjz: PTAL.

[xjz, 2016-06-29 23:46:53]

https://codereview.chromium.org/2108373002/diff/1/media/cast/sender/external_...
File media/cast/sender/external_video_encoder.cc (right):

https://codereview.chromium.org/2108373002/diff/1/media/cast/sender/external_...
media/cast/sender/external_video_encoder.cc:47: // "deadline utilization"
heuristic to measure encoder utilization. Otherwise,
s/deadline/encoder

https://codereview.chromium.org/2108373002/diff/1/media/cast/sender/external_...
media/cast/sender/external_video_encoder.cc:53: //  
--cast-encoder-util-heuristic=foobar    ==> 0  (unrecognized, use deadline)
s/deadline/encoding time?

https://codereview.chromium.org/2108373002/diff/1/media/cast/sender/external_...
media/cast/sender/external_video_encoder.cc:359: backlog_redline_threshold_;
Shall we use |in_progress_frame_encodes_.size()-1| here since the current frame
is already done with encoding?

[miu, 2016-06-30 20:38:45]

https://codereview.chromium.org/2108373002/diff/1/media/cast/sender/external_...
File media/cast/sender/external_video_encoder.cc (right):

https://codereview.chromium.org/2108373002/diff/1/media/cast/sender/external_...
media/cast/sender/external_video_encoder.cc:47: // "deadline utilization"
heuristic to measure encoder utilization. Otherwise,
On 2016/06/29 23:46:52, xjz wrote:
> s/deadline/encoder

No, it's correct the way it is. Here, I am saying that a return value of 0 means
we should use the old heuristic, which is the deadline utilization computation.

https://codereview.chromium.org/2108373002/diff/1/media/cast/sender/external_...
media/cast/sender/external_video_encoder.cc:53: //  
--cast-encoder-util-heuristic=foobar    ==> 0  (unrecognized, use deadline)
On 2016/06/29 23:46:52, xjz wrote:
> s/deadline/encoding time? 

Same here. 0 means use the deadline computation.

https://codereview.chromium.org/2108373002/diff/1/media/cast/sender/external_...
media/cast/sender/external_video_encoder.cc:359: backlog_redline_threshold_;
On 2016/06/29 23:46:52, xjz wrote:
> Shall we use |in_progress_frame_encodes_.size()-1| here since the current
frame
> is already done with encoding?

By encoding the frame, we utilized the encoder in some greater-than-zero amount.
So, the current frame needs to be included in the calculation.

Another way to look at this: If the current frame is the only one in backlog, we
still need to report a utilization greater than zero.

BTW--I took another look at this, and added a bit more explanation about our
model/thinking on this problem in the code comments here.

[xjz, 2016-06-30 20:51:33]

lgtm

[chcunningham, 2016-06-30 21:12:26]

media lgtm

[no sievers, 2016-06-30 21:50:24]

On 2016/06/30 21:12:26, chcunningham wrote:
> media lgtm

[no sievers, 2016-06-30 21:50:25]

On 2016/06/30 21:12:26, chcunningham wrote:
> media lgtm

[no sievers, 2016-06-30 21:50:39]

content lgtm

[miu, 2016-06-30 23:04:27]

The CQ bit was checked by miu@chromium.org

[miu, 2016-06-30 23:04:27]

The patchset sent to the CQ was uploaded after l-g-t-m from xjz@chromium.org,
chcunningham@chromium.org, sievers@chromium.org
Link to the patchset: https://codereview.chromium.org/2108373002/#ps40001
(title: "REBASE")

[commit-bot: I haz the power, 2016-06-30 23:04:58]

CQ is trying da patch. Follow status at
 
https://chromium-cq-status.appspot.com/v2/patch-status/codereview.chromium.or...

[commit-bot: I haz the power, 2016-07-01 01:00:31]

Description was changed from

==========
Add experimental 'backlog' heuristic for measuring encoder utilization.

The existing 'deadline' heuristic has proven to be very flawed on some
systems, for reasons explained in the bug. This change adds a new
experimental heuristic, activated via a new command-line switch, for lab
testing: --cast-encoder-util-heuristic=backlog<N>

If the new heuristic proves to work well across our lab's gamut of test
systems, a future change will remove the command-line switch and make
the new heuristic the default.

BUG=chrome-os-partner:54806
==========

to

==========
Add experimental 'backlog' heuristic for measuring encoder utilization.

The existing 'deadline' heuristic has proven to be very flawed on some
systems, for reasons explained in the bug. This change adds a new
experimental heuristic, activated via a new command-line switch, for lab
testing: --cast-encoder-util-heuristic=backlog<N>

If the new heuristic proves to work well across our lab's gamut of test
systems, a future change will remove the command-line switch and make
the new heuristic the default.

BUG=chrome-os-partner:54806
==========

[commit-bot: I haz the power, 2016-07-01 01:00:33]

Committed patchset #3 (id:40001)

[commit-bot: I haz the power, 2016-07-01 01:02:40]

Description was changed from

==========
Add experimental 'backlog' heuristic for measuring encoder utilization.

The existing 'deadline' heuristic has proven to be very flawed on some
systems, for reasons explained in the bug. This change adds a new
experimental heuristic, activated via a new command-line switch, for lab
testing: --cast-encoder-util-heuristic=backlog<N>

If the new heuristic proves to work well across our lab's gamut of test
systems, a future change will remove the command-line switch and make
the new heuristic the default.

BUG=chrome-os-partner:54806
==========

to

==========
Add experimental 'backlog' heuristic for measuring encoder utilization.

The existing 'deadline' heuristic has proven to be very flawed on some
systems, for reasons explained in the bug. This change adds a new
experimental heuristic, activated via a new command-line switch, for lab
testing: --cast-encoder-util-heuristic=backlog<N>

If the new heuristic proves to work well across our lab's gamut of test
systems, a future change will remove the command-line switch and make
the new heuristic the default.

BUG=chrome-os-partner:54806

Committed: https://crrev.com/70c68b6a6eb9a3c46662b5ed112f36d53f8e9813
Cr-Commit-Position: refs/heads/master@{#403378}
==========

[commit-bot: I haz the power, 2016-07-01 01:02:41]

Patchset 3 (id:??) landed as
https://crrev.com/70c68b6a6eb9a3c46662b5ed112f36d53f8e9813
Cr-Commit-Position: refs/heads/master@{#403378}