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 6b69d15098bbb3cd6b13596f615eeb8bff1cef59..7d9fad4b8fd26e7ee590b87d5bfd344530ed61fe 100644
--- a/tools/metrics/histograms/histograms.xml
+++ b/tools/metrics/histograms/histograms.xml
@@ -13714,6 +13714,44 @@ other types of suffix sets.
   </summary>
 </histogram>
 
+<histogram name="Precache.DailyDownloadedNonPrecacheKB" units="KB">
+  <summary>
+    The number of kilobytes that were downloaded over the network for HTTP/HTTPS
+    fetches that were not motivated by precaching. This metric is accumulated
+    over 24 hour intervals, and can cover multiple sessions. Values of this

[jar (doing other things), 2013/11/26 20:52:53]

FYI: This aggregation over 24 hours will likely conflate the length of time that
a user runs (without restart), with the data I think you really wanted.

I suspect you wanted to measure the size of each resource separately, and then
aggregate the histograms.  That would remove the conflation I mentioned above.

[sclittle, 2013/11/26 22:22:28]

I've updated the descriptions to hopefully help clarify that this is
per-profile.

We want to measure the usefulness and overhead of precaching for the typical
user. If we measure the savings and overhead of each resource separately, this
would show us the savings and overhead for the typical resource, which we don't
care about. Aggregating per-resource histograms also wouldn't tell us much about
typical users either, since those sums would be biased toward outliers who
fetched a lot of big resources. By aggregating metrics over 24 hours for each
profile, we can measure things like the median user's savings percentage.

+    metric for previous, unreported 24 hour intervals are reported at the start
+    of a precaching cycle.
+  </summary>
+</histogram>
+
+<histogram name="Precache.DailyDownloadedPrecacheMotivatedKB" units="KB">
+  <summary>
+    The number of kilobytes that were downloaded because of precaching. This
+    metric is accumulated over 24 hour intervals, and can cover multiple
+    sessions. Values of this metric for previous, unreported 24 hour intervals
+    are reported at the start of a precaching cycle.
+  </summary>
+</histogram>
+
+<histogram name="Precache.DailySavedKB" units="KB">
+  <summary>
+    The number of kilobytes during user browsing that were served from the
+    cache, but would have been downloaded over a network if precaching was
+    disabled. This metric is accumulated over 24 hour intervals, and can cover
+    multiple sessions. Values of this metric for previous, unreported 24 hour
+    intervals are reported at the start of a precaching cycle.
+  </summary>
+</histogram>
+
+<histogram name="Precache.DailySavingsPercentage" units="%">
+  <summary>
+    The percentage of user browsing network bandwidth that was saved because of
+    precaching. This metric is accumulated over 24 hour intervals, and can cover
+    multiple sessions. Values of this metric for previous, unreported 24 hour
+    intervals are reported at the start of a precaching cycle.
+  </summary>
+</histogram>
+
 <histogram name="Prerender.Events" enum="PrerenderHoverEvent">
   <obsolete>
     deprecated May 10 2012
@@ -31507,6 +31545,14 @@ other types of suffix sets.
   <affected-histogram name="Plugin.PpapiPluginLoadResult"/>
 </fieldtrial>
 
+<fieldtrial name="PrecacheCellular" separator=".">
+  <group name="Cellular"
+      label="covers fetches when connected to cellular networks"/>
+  <affected-histogram name="Precache.DailyDownloadedNonPrecacheKB"/>
+  <affected-histogram name="Precache.DailySavedKB"/>
+  <affected-histogram name="Precache.DailySavingsPercentage"/>
+</fieldtrial>
+
 <fieldtrial name="Prefetch">
   <group name="ContentPrefetchPrefetchOff"
       label="Prefetch is completely disabled."/>