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 488395dfcb1c70d3baf34e5ef734e0faacd8eede..eff171e39b729ddd17002c9e192e6ad01cb6ef83 100644
--- a/tools/metrics/histograms/histograms.xml
+++ b/tools/metrics/histograms/histograms.xml
@@ -2701,6 +2701,16 @@ Therefore, the affected-histogram name has to have at least one dot in it.
   </summary>
 </histogram>
 
+<histogram name="ChromeOS.MemoryPressureHandleQuality.NonCriticalRatio">
+  <owner>xdai@chromium.org</owner>
+  <summary>
+    A by 10000% multiplied fraction of time spent in non critical vs. non
+    critical + critical time. This can be used to compare different memory
+    pressure handling schemes against each other to evaluate whether a higher
+    number is better.

[Ilya Sherman, 2015/02/02 19:55:50]

Hmm, from looking at this description, I only partially understand this metric. 
I think it might help in part to write the ratio as a ratio, with parentheses
enclosing the numerator and denominator, like so: (non-critical) / (non-critical
+ critical).  Then, you say "10000%" -- which is equal to 100 -- but it looks
like the code multiplies by 1 million.  I'd recommend writing this without a
percent, as that just makes the value confusing.  I'd also mention *why* you're
scaling the number -- I assume it's because UMA can only record integers, so
this is how you're representing the fractional values that you actually care
about.

Now, all that said, I see that you're using an exponentially-spaced histogram,
i.e. a histogram that has higher resolution for smaller values, and relatively
low resolution for larger values.  Is that appropriate for the metric that
you're computing?  Also, multiplying by 1 million suggests that you believe that
you have at least 6 significant figures in the measurement.  Is that true? 
That's pretty surprising to me -- I would have expected something closer to two
significant digits.

In addition, it's not clear to me how to interpret the data in this histogram --
probably because I don't wok on memory pressure.  It would be helpful for the
reader to know whether higher or lower non-critical ratios are better; and
ideally a little bit of insight as to why this ratio, in particular, is a useful
one to measure.

Finally, please document when this metric is recorded.  I'm guessing it's either
periodic -- say once every minute -- or triggered by some event -- maybe process
termination?  But I'm really not sure.

+  </summary>
+</histogram>
+
 <histogram name="ChromeOS.PlatformVerification.Available"
     enum="BooleanAvailable">
   <owner>dkrahn@chromium.org</owner>