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 00e5a9355a0a00fa55c17adb3ba7314ff0a3952f..49d5d9a7a77638603cdf5570d4fbfe83501cc99c 100644
--- a/tools/metrics/histograms/histograms.xml
+++ b/tools/metrics/histograms/histograms.xml
@@ -2602,13 +2602,25 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
 <histogram name="Bookmarks.LaunchDepth">
   <owner>yfriedman@chromium.org</owner>
   <summary>
-    Logs the depth of the bookmark in the bookmark tree hiearchy every time a
+    Logs the depth of the bookmark in the bookmark tree hierarchy every time a
     bookmark is launched. Depth indicates how many levels below a permanent
     bookmark folder the bookmark was found in (e.g. a bookmark immediately in
     the bookmark bar has depth 1).
   </summary>
 </histogram>
 
+<histogram name="Bookmarks.OnFaviconsChangedIconURL" enum="BooleanHit">
+  <owner>pkotwicz@chromium.org</owner>
+  <summary>
+    Logs how frequently BookmarkModel::OnFaviconsChanged() is called with a
+    non-empty |icon_url| to determine whether BookmarkNodes should be cached
+    based on their favicon URLs (e.g. http://www.google.com/favicon.ico) in
+    addition to their page URLs (e.g. http://www.google.com). OnFaviconChanged()
+    is expected to be called less than 10 times per user per day with a
+    non-empty |icon_url|.

[Ilya Sherman, 2015/07/06 17:57:40]

How do you plan to interpret the numbers in this histogram?

[pkotwicz, 2015/07/06 18:47:30]

I am planning on figuring out the average # of times that this histogram gets
logged per user per day. I may need to get logs access for that. If the average
is > 10 per day, then I will consider adding caching for BookmarkNodes based on
icon URL

[Ilya Sherman, 2015/07/07 02:40:36]

Hmm, I'd discourage you from just looking at an average.  For example, suppose
that 50% of users log this histogram once per day, and 50% log it 17 times per
day.  That looks like 9 times per day on average; but really, there are two
types of users: ones who hit it rarely, and ones who hit it frequently. 
Alternately, suppose that 99% of users log this histogram once per day, and a
very atypical 1% of users log it 1000 times per day -- perhaps because they are
actually page cycler bots.  Again, you'd get misleading data from this sort of
usage pattern.

Rather than looking at an average, I'd recommend trying to find a way that gives
you a per-user score or "cost", so that you can look at a distribution of costs
for analysis.  If you're willing to write a Dremel script, then this histogram
structure will probably give you the information that you need, so LGTM.  But I
do strongly recommend taking this advice into account when conducting the
statistical analysis of the data.

+  </summary>
+</histogram>
+
 <histogram name="BrowserRenderProcessHost.ChildCrashes" enum="RendererType">
   <owner>wfh@chromium.org</owner>
   <summary>Count of renderer process crashes grouped by type.</summary>