Update loading_measurement_analyzer to display CPU vs Network breakdowns.

Update loading_measurement_analyzer to display CPU vs Network breakdowns.

BUG=None
NOTRY=True

Committed: https://src.chromium.org/viewvc/chrome?view=rev&revision=213293

[dtu, 2013-07-22 22:40:41]

lgtm

l

[nduca, 2013-07-22 23:37:00]

hmm

https://codereview.chromium.org/19925002/diff/1/tools/perf/measurements/loadi...
File tools/perf/measurements/loading_measurement_analyzer.py (right):

https://codereview.chromium.org/19925002/diff/1/tools/perf/measurements/loadi...
tools/perf/measurements/loading_measurement_analyzer.py:64:
self.totals[key].append((value, row['url']))
so you define network time as the unaccounted-for time?

what happens when you have this?
  Begin ts=10 Foo
  Begin ts=15 Bar
  End    ts=20 Bar
  End    ts=25 Foo

This gives us foo duration=15 Bar duration=5. Cpu time consumed is actually 15
in this case...

[tonyg, 2013-07-23 02:23:21]

On 2013/07/22 23:37:00, nduca wrote:
> hmm
> 
>
https://codereview.chromium.org/19925002/diff/1/tools/perf/measurements/loadi...
> File tools/perf/measurements/loading_measurement_analyzer.py (right):
> 
>
https://codereview.chromium.org/19925002/diff/1/tools/perf/measurements/loadi...
> tools/perf/measurements/loading_measurement_analyzer.py:64:
> self.totals[key].append((value, row['url']))
> so you define network time as the unaccounted-for time?
> 
> what happens when you have this?
>   Begin ts=10 Foo
>   Begin ts=15 Bar
>   End    ts=20 Bar
>   End    ts=25 Foo
> 
> This gives us foo duration=15 Bar duration=5. Cpu time consumed is actually 15
> in this case...

That isn't quite what I'm doing. Here's what I'm doing:

navigationStart= 0
Begin ts=10 Foo
Begin ts=15 Bar
End    ts=20 Bar
End    ts=25 Foo
loadEventStart=35

Page load time = 35
CPU time = 15
Network time = 20

This works with timeline events because the Program event covers all main thread
time. It is instrumented right at the message loop level. To verify, I ran an
experiment with unlimited network and the cpu time was roughly 100% of the page
load time. So I'm attributing the network time as the amount of time the page
load time would have decreased by had the network been unlimited.

Maybe this assumption doesn't hold true for the trace based measurement?

[nduca, 2013-07-23 03:00:26]

Agh, I got it. I forgot this was all self_time based! Fail on my part, my
apologies.

Long term, I think devtools actually issues events [that we ignore] for when the
message loop is idle. Might be a good followup task to use those to construct an
"OTHER" category so that if there's uninstrumented bits that inspector doesn't
catch, we'd catch it. But, followup.

[nduca, 2013-07-23 03:00:50]

(lgtm)

[tonyg, 2013-07-23 03:30:26]

On 2013/07/23 03:00:26, nduca wrote:
> Agh, I got it. I forgot this was all self_time based! Fail on my part, my
> apologies.
> 
> Long term, I think devtools actually issues events [that we ignore] for when
the
> message loop is idle. Might be a good followup task to use those to construct
an
> "OTHER" category so that if there's uninstrumented bits that inspector doesn't
> catch, we'd catch it. But, followup.

The self-time of "Program" is "OTHER".

[nduca, 2013-07-23 03:38:45]

yeahhh i'm sorry for the idiocy. too much talk today. :)

[commit-bot: I haz the power, 2013-07-24 00:06:12]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/tonyg@chromium.org/19925002/1

[commit-bot: I haz the power, 2013-07-24 00:28:04]

Change committed as 213293