Benchmark to characterize main thread usage during load

Benchmark to characterize main thread usage during load

This uses inspector timeline and telemetry to get an estimate of time consumption during page load.

NOTRY=True

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=176132

[nduca, 2013-01-08 06:35:21]

just a rough sketch. I think the data is wrong but the approach is roughly how I
think we'd do it in telemetry.

A key thing in telemetry is we are trying to separate out "what to with the
page" from how to measure it. The blahblah_benchmark.py is supposed to do the
measuring, while the .json file is supposed to just say what to do to a page to
achieve a given result.

Unfortunately, we're in the midst of this separation, so really only the
page_sets/top_25.json really conforms to this separation. We'll get better over
the next few weeks at this, the timing of this discussion is just a bit early.
:)

[nduca, 2013-01-08 07:00:24]

Ah, the data set is funky because some inspector events dont actually have end
times.

[nduca, 2013-01-08 13:12:42]

patchset 3 depends on https://codereview.chromium.org/11779041/, which should
land tomorrow to clean up csv output logic.

[nduca, 2013-01-09 21:38:47]

Eric, in theory this runs ToT. Maybe you can pick it up from here and put up a
patch for proper landing?

[abarth-chromium, 2013-01-09 21:40:40]

On 2013/01/09 21:38:47, nduca wrote:
> Eric, in theory this runs ToT. Maybe you can pick it up from here and put up a
> patch for proper landing?

Thanks for updating the patch.  I'll give it a try as well.  :)

[nduca, 2013-01-09 21:44:01]

Needs chrome >= r175886

[abarth-chromium, 2013-01-10 01:51:29]

This works great!  LGTM, but I'm probably not an owner.

[nduca, 2013-01-10 10:42:08]

Hows this as a landable benchmark? I've fixed most of the issues.

An open is the name: Webkit didn't make sense as a benchmark because it was
useful for other kinds of things as well.

Given the direction dtu is heading with crbug.com/168431, I think our benchmarks
will be organized into "stress memory", "smoothness" and "loading." So calling
this loading felt like it maybe fit?

[tonyg, 2013-01-10 18:53:05]

lgtm with nit

This is awesome!

https://codereview.chromium.org/11791043/diff/8002/tools/perf/perf_tools/load...
File tools/perf/perf_tools/loading_benchmark.py (right):

https://codereview.chromium.org/11791043/diff/8002/tools/perf/perf_tools/load...
tools/perf/perf_tools/loading_benchmark.py:26: if e.name not in events_by_name:
Use collections.defaultdict(list) to clean this up.

[nduca, 2013-01-10 18:55:32]

> Use collections.defaultdict(list) to clean this up.
Holy crap thats cool. I learned something new about python today!

[nduca, 2013-01-10 19:39:15]

(note to sheriffs: notry=true because none of this code is live on the main
waterfall)

[commit-bot: I haz the power, 2013-01-10 19:39:59]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/nduca@chromium.org/11791043/15003

[commit-bot: I haz the power, 2013-01-10 19:43:07]

Change committed as 176132