tools/metrics/histograms/histograms.xml

Index: tools/metrics/histograms/histograms.xml
diff --git a/tools/metrics/histograms/histograms.xml b/tools/metrics/histograms/histograms.xml
index a8a09984d5a4551be58098b33f31c15489744766..be5d4c6ffeff41fb0b33e0bc3563d0f95bc459c0 100644
--- a/tools/metrics/histograms/histograms.xml
+++ b/tools/metrics/histograms/histograms.xml
@@ -11315,6 +11315,17 @@ other types of suffix sets.
   </summary>
 </histogram>
 
+<histogram name="PerformanceMonitor.AverageCPU.BrowserProcess">
+  <summary>
+    Average CPU utilization of the browser process, read out at each two-minute
+    interval. On Windows this is a range from 0-100% of the total CPU capacity,
+    i.e. a quadcore system with one core running fully loaded and the other
+    three idle would read as 25%. On non-Windows systems there's a 0-100% range
+    per CPU, which is then summed up. I.e. a quadcore system fully loaded would
+    read as 400%.

[Alexei Svitkine (slow), 2013/10/21 14:53:46]

Can we perhaps transform the value such that its in the same range for all
platforms?

i.e. either multiply the value by the number of cores/cpus on Windows or divide
it on other platforms?

[oystein (OOO til 10th of July), 2013/10/21 15:15:06]

The plan is to eventually de-normalize the Windows one yeah (rather than the
other way around, to avoid losing information for the non-Windows platforms).
Since that involves extending some of the base libraries I didn't really want to
bundle those changes with this, but I'm happy to do it if you think it's
important to keep the values consistent between platforms.

[Alexei Svitkine (slow), 2013/10/21 15:27:31]

I think it's useful to keep it consistent from the start. Otherwise, when you do
make that change, you'll need to add a new histogram (obsoleting this one) that
has the new semantic.

[oystein (OOO til 10th of July), 2013/10/21 15:33:12]

Oh, really? Even if the ranges and buckets stay the same? I was assuming the
change would effectively invalidate the historical Windows data.

[Alexei Svitkine (slow), 2013/10/21 16:31:07]

Right, the problem is it's not really possible to invalidate the old data (i.e.
in the system). For example, if someone were to aggregate the histogram values
(i.e. select the "Everything" version), then you'd get the old values aggregated
with the new ones, leading to incorrect results. For this reason, we suggest any
semantic change to a histogram should rename it (i.e. create a new one and
obsolete the old one in histograms.xml).

This is why it's better to make the other change to standardize the values
first.

[oystein (OOO til 10th of July), 2013/10/21 17:30:03]

Sure thing. The main reason I haven't already done this is lack of access to a
Windows machine (won't have that for a few weeks), but I'll figure something out
:).

+  </summary>
+</histogram>
+
 <histogram name="PerformanceMonitor.HighCPU.BrowserProcess" enum="BooleanHit">
   <summary>
     The number of times a browser process has continuously stayed above a