Telemetry: adds CPU frequency stats.

tools/perf/unit-info.json

Index: tools/perf/unit-info.json
diff --git a/tools/perf/unit-info.json b/tools/perf/unit-info.json
index 5ace7830efb16a0165e45b30e5b24489fbd870a4..15006ac1f094692554dcc1b895b65759a822c0c4 100644
--- a/tools/perf/unit-info.json
+++ b/tools/perf/unit-info.json
@@ -81,6 +81,10 @@
   "MB": {
     "improvement_direction": "down"
   },
+  "MHz": {
+    "improvement_direction": "down",
+    "why": "CPU governors were able to keep the CPU at lower frequency."

[pasko, 2014/04/23 15:18:54]

I thought we don't need to track regressions for this metric just yet, but
rather want to find correlation with power draw. We are not enabling the alerts
for this metric in near future. In other words this adds some 'dead code', not
covered by tests.

Truly hypothetically: this may backfire in the future if someone aims at
maximizing MHz during a certain action to save power (make critical computations
faster). Will that reverse the improvement_direction for MHz? This example is
not something realistic, but rather demonstrates my feelings towards introducing
the unnecessary.

[qyearsley, 2014/04/23 18:38:22]

Fair enough. This unit-info.json is only used by the perf dashboard; if no entry
for MHz was added here, then on the result graphs, on the Y-axis it would
display: "MHz (? is better)"

[bulach, 2014/05/06 17:08:19]

removed this file..

+  },
   "mips": {
     "improvement_direction": "up",
     "why": "More instructions processed per time unit."