All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=847440

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=b5121e61817a3f09984e0028f443b1f7b24e2d844757eebbf207fa2b58b6a311


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

linux-perf

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/11c28b46240000

📍 Found significant differences after each of 9 commits.
https://pinpoint-dot-chromeperf.appspot.com/job/11c28b46240000

Add LevelDB implementation of DownloadDB by qinmin@chromium.org
https://chromium.googlesource.com/chromium/src/+/826f29b4b11aaed28c34cacf62087d703515a4f3
1986 → 2006 (+20.24)

[dataview] Move DataView getters and setters to Torque by theotime@google.com
https://chromium.googlesource.com/v8/v8/+/c2614f9f7db48a39253fbb91322d95268825599e
2001 → 2015 (+13.08)

[runtime] Do not shrink fixed arrays to length 0. by neis@chromium.org
https://chromium.googlesource.com/v8/v8/+/5a0ebc8ebc42dc9fbad14cdf13e336d4e1be001d
2010 → 2026 (+15.47)

[typedarray] Change TypedArray.p.sort implementation. by szuend@google.com
https://chromium.googlesource.com/v8/v8/+/a6c71508c5a026a299255333f3152234c6d98914
2028 → 2071 (+42.63)

[in-place weak refs] Replace the WeakCell(transition_map) in DataHandlers. by marja@chromium.org
https://chromium.googlesource.com/v8/v8/+/99bf906887bacb9c9bcdb83d030ff748f0bbb19a
2052 → 2031 (-21.43)

[wasm] Emit IA disassembly for --trace-turbo by herhut@chromium.org
https://chromium.googlesource.com/v8/v8/+/ee82333bc8b35a6517bc81785a31fb228fffe7cd
2031 → 2018 (-12.24)

[wasm] Avoid embedding {null} values in WasmCode. by mstarzinger@chromium.org
https://chromium.googlesource.com/v8/v8/+/fabb5140876dbc1eb17b03c2b46b2f552bba2e0a
2018 → 2032 (+13.93)

[builtins] Move ArrayConstructorStub to builtin by jgruber@chromium.org
https://chromium.googlesource.com/v8/v8/+/111c5735efde1efa68b37e493a5f7517ae3712c8
2021 → 2005 (-16.51)

Roll AFDO from 68.0.3440.0_rc-r1 to 69.0.3441.0_rc-r1 by afdo-chromium-autoroll@skia-buildbots.google.com.iam.gserviceaccount.com
https://chromium.googlesource.com/chromium/src/+/bf20d87614794a2ec083cdebff80a4709e79e8db
2003 → 2263 (+259.5)

Assigning to sheriff gbiv@chromium.org because "Roll AFDO from 68.0.3440.0_rc-r1 to 69.0.3441.0_rc-r1" is a roll.

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions

Interesting; I'll buy that the AFDO roll is the culprit.

For background, AFDO works by sampling Chrome's execution and feeding those samples back into the compiler, so the compiler can make better optimization decisions. Since it's built around periodic sampling, there's a fair amount of noise in this process. Blink benchmarks are particularly good at highlighting this noise, since they tend to be very sensitive to individual optimization decisions. :)

In fewer words, this may (read: will hopefully) magically disappear after a few more rolls.

In general, we're seeing a fair amount of noise lately related to AFDO rolls; my goal is to reduce that, so I'll carry this bug while I do that.

Automatic message: All alerts recovered.
Graphs: https://chromeperf.appspot.com/group_report?bug_id=847440