All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=754609

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=145a2444ae58c0cfbbb6e3fa12be8280acb33c6fd244efd5c5265f6f48a0bc3b


Bot(s) for this bug's original alert(s):

webrtc-android-tests-nexus5-kitkat

Hey Edward,

As you can see in several of the graphs in the attached report, the Nexus 5 perf bot seems to oscillate between two "performance levels". This is particularly visible in WebRTCPerf/webrtc-android-tests-nexus5-kitkat/webrtc_perf_tests / dropped_frames / simulcast_vp8_3sl_high.

Could this be an infra problem, maybe due to the tests running on different devices at different times?

Yes, that might have happened.
The bot has 6 devices, and if devices die or come back alive, the distribution of the tests will change.
I'll take a look.

Thanks for looking into it. I've added some more alerts to the report.

Since these are all related to the same device type (nexus 5), I'm pretty sure it must be device availability problems.

It looks pretty stable now so I'm closing.