Electrolysis/Release Criteria/UI Smoothness

From MozillaWiki
Jump to: navigation, search

Some UI smoothness metrics may regress, on Windows and Mac, up to 5%, if that gets wins elsewhere.


  • Responsible: mconley
  • Accountable: RyanVM
  • Supporting: jmaher, avih, gabor
  • Consulted:
  • Informed: cpeterson, elan, release management


talos TART (tab animation smoothness)

  • e10s comparison validated: mconley
  • signed off by barbara over email on April 20th, 2016
  • Current e10s diff: No worse than non-e10s with APZ disabled (see this analysis)
  • bug 1251699 - 5.7% TART regression on OS X with APZ disabled
    • Closed as WORKSFORME - this regression seems to have disappeared.
  • bug 1253412 - APZ causes 4% - 12% TART regression - Event Region Regression

talos tpaint (first paint of a new window opened from content)

  • e10s comparison validated: jimm
  • Current e10s diff: worse by 3-8% on Linux 64, Windows 7 and Windows 8 platforms, with APZ on or off. 11% improvement on OS X
  • bug 1174770 - tpaint regressions (3%-8%) in e10s mode compared to non-e10s mode


  • tpaint measures the time it takes to paint the content for a new window that has been opened from content. Windows can be opened from content in a few ways:
    • The content runs some JavaScript that calls window.open with non-default features, like window.open("https://www.mozilla.org", "_blank", "toolbars=no");
    • The user has flipped the browser.link.open_newwindow pref from its default to open new windows instead of new tabs, and then clicks on a target="_blank" anchor
  • tpaint does not measure the time it takes to open a window triggered from the parent. Windows can be opened from the parent in a number of ways:
    • The user uses Ctrl-N or File > New Window to open a new window
    • The user clicks on a link in another application that opens the browser the first time
    • The user right clicks on a link in content, and chooses to Open Link in New Window
    • The user tears a tab out into a new window
  • An experiment was conducted to determine whether or not e10s regresses opening windows from the parent.
    • The experiment was conducted in bug 1251732
    • The conclusion was that e10s does not regress this case.
  • We suspect there is a second slow-path here that Talos is not measuring, and that's when the user clicks on a link that opens a new tab.
    • A bug has been filed to add such a test (bug 1253382)
    • Preliminary local results from that test show that e10s does not seem to regress this case
  • A bug has been filed to get Telemetry on how often we hit the ContentParent::RecvCreateWindow codepath, which we believe is the slow path. (bug 1253388)

talos tresize (window resize responsiveness)

  • e10s comparison validated: jimm
  • Current e10s diff: mixed
    • Linux 64: ~7% regression
    • OS X: ~2% regression
    • Windows 7: ~2% regression
    • Windows 8 64: ~17% improvement
    • Windows XP: ~9% regression
  • Signed off by canuckistani here.
  • bug 1179732 - 11-147% linux*/winxp tresize regression when comparing e10s to non e10s on the same build
  • bug 1252947 - ~5% winxp tresize regression when comparing e10s to non e10s on the same build

talos tps (tab-switch speed)

  • e10s comparison validated: jimm
  • Current e10s diff:
    • Linux 64: 19.38% improvement
    • OS X: 35.59% improvement
    • Win 7: 26.61% improvement
    • Win 8: 33.78% improvement
    • Win XP: 26.26% improvement

talos CART (customization mode animation test)

  • e10s comparison validated: mconley
  • Current e10s diff (with APZ disabled - see this analysis)
    • Win 8: 2.0% worse
    • Win 7: 4.0% worse
    • OS X: 3.7% worse
    • Linux 64: 6.8% worse
  • bug 1252768 - 2–7% e10s CART regression
    • The regressions on non-Linux platforms is due to the "New non-e10s window" button that is added to Nightly and Aurora builds
    • The regression for Linux 64 is not yet understood.


Full Query
ID Summary Whiteboard Keywords Assigned to
1252768 2–7% e10s CART regression (with APZ disabled) perf
1253388 Add a Telemetry probe to see how often we go through RecvCreateWindow to create a new window or tab btpp-fixlater

2 Total; 2 Open (100%); 0 Resolved (0%); 0 Verified (0%);