1096756 - e10s very slow to start up

Status: RESOLVED FIXED

(Firefox :: Untriaged, defect)

Description

[Nikolay Martynov]

User Agent: Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:36.0) Gecko/20100101 Firefox/36.0
Build ID: 20141110182958

Steps to reproduce:

Today's firefox-trunk has e10s enabled by default. First thing that I've noticed is that start up of my session with two windows and about 20-30 tabs is very slow.

I use ctrl+pageup to 'activate' all tabs I have.


Actual results:

Loading of the tabs seems to take very long time. Much-much-much longer then with e10s disabled. Firefox is consuming lots of CPU during this.

(And somewhere in the middle of the loading process #1096754 happens)


Expected results:

Loading of tabs with e10s enabled should not be slower than with e10s disabled.

Comment 1

[Mike Conley (:mconley) (:⚙️)]

We suspect this will be better now that bug 1096013 has landed. Do you find restore performance has improved with an up-to-date Nightly?

Comment 2

[Nikolay Martynov]

I will retest it once new nightly build appears for ubuntu on ppa

Comment 3

[Vladan Djeric (:vladan)]

There's a noticeable regression in startup time in the Telemetry data as well: bug 1099406

Comment 4

[Nikolay Martynov]

Ok, I've given today's trunk a try and can say the following:

1) It still takes much longer time to load all tabs that I have. And firefox locks often during this process so I cannot really measure any timings. I've files a separate bug about lock ups.

2) I have two windows in my session and when I start firefox both of them pop up (expected) and wuth e10s enabled I see large 'spinner' instead of the page in both windows (expected, I guess). The problem is that often one of the windows stays in that tab forever. I.e. it never actually displays page. I can start switching tabs in that window and this seem to make it work. But if I touch nothing after I start firefox up default tab in one of the windows never loads. This kind of adds to the feeling that loading is very slow.

3) It looks like firefox is less 'interactive' when tabs are loading. 'Interactive' might be not exactly correct word though, I'll try to explain better.
   Without e10s when I when load many tabs at the same time UI becomes laggy. But when one tab has finished loading it becomes usable (e.g. gmail) - I can switch to other (loading) tabs and back to gmail quickly - loading tabs are still loading, gmail is usable. This is good, because even though my 20+ tabs are still loading I can already use gmail, and it looks like firefox gives some preference to 'active' tab so which ever tab I look at loads quickly.
   With e10s enabled situation is different. When many tabs are loading I see large 'spinner'. Eventually active tab displays page. But then if I switch to any other tab and back to the tab that has already loaded I see 'spinner' again for at least several seconds, some times for a long time. This is not good - this means that if I have many tabs loadin gin background I cannot really switch between tabs and use already loaded tabs.
   This feels like there is only one thread that loads all tabs and it becomes busy with tabs that are loading on background and takes very long time respond with content for the tab that has already been loaded - I'm not pretending firefox does this, I'm just trying to provide an explanation that would illustrate thing that I'm seeing. Hope this makes sense.

  Please let me know if I can provide any more information or if I should file separate bugs for things that I've outlined about.

Thanks!

Comment 5

[Chris Peterson [:cpeterson]]

Nikolay, do you have the "browser.sessionstore.restore_on_demand" about:config pref set to false?

Can you please copy/paste your about:support configuration information? Go to Help menu > "Troubleshooting Information" menu item > "Copy text to clipboard" button.

Comment 6

[Nikolay Martynov]

restore_on_demand is true, and it is true by default as far as I can see. Just a side note: when I start FF up I tend to crtl+pgup though all tabs to start load process which I guess basically negates that setting.

Troubleshooting information:

Application Basics
------------------

Name: Firefox
Version: 36.0a1
User Agent: Mozilla/5.0 (X11; Ubuntu; Linux x86_64; rv:36.0) Gecko/20100101 Firefox/36.0
Multiprocess Windows: 0/4

Extensions
----------

Name: Adblock Plus
Version: 2.6.6
Enabled: true
ID: {d10d0bf8-f5b5-c8b4-a8b2-2b9879e08c5d}

Name: Firebug
Version: 2.0.6
Enabled: true
ID: [e-mail]

Name: Greasemonkey
Version: 2.3
Enabled: true
ID: {e4a8a97b-f2ed-450b-b12d-ee082ba24781}

Name: gTranslate
Version: 0.9
Enabled: true
ID: {aff87fa2-a58e-4edd-b852-0a20203c1e17}

Name: NoSquint
Version: 2.1.9
Enabled: true
ID: [e-mail]

Name: Secure Password Generator
Version: 0.5
Enabled: true
ID: [e-mail]

Name: UAControl
Version: 0.1.3.1
Enabled: true
ID: [e-mail]

Name: Web Developer
Version: 1.2.5
Enabled: true
ID: {c45c406e-ab73-11d8-be73-000a95be3b12}

Name: Flashblock
Version: 1.5.17
Enabled: false
ID: {3d7eb24f-2740-49df-8937-200b1cc08f8a}

Graphics
--------

Adapter Description: Intel Open Source Technology Center -- Mesa DRI Intel(R) Ivybridge Mobile
Device ID: Mesa DRI Intel(R) Ivybridge Mobile
Driver Version: 3.0 Mesa 10.4.0-devel (git-920f875 2014-11-17 trusty-oibaf-ppa)
GPU Accelerated Windows: 4/4 OpenGL (OMTC)
Vendor ID: Intel Open Source Technology Center
WebGL Renderer: Intel Open Source Technology Center -- Mesa DRI Intel(R) Ivybridge Mobile
windowLayerManagerRemote: true
AzureCanvasBackend: skia
AzureContentBackend: skia
AzureFallbackCanvasBackend: none
AzureSkiaAccelerated: 0

Important Modified Preferences
------------------------------

accessibility.typeaheadfind.flashBar: 0
browser.cache.disk.capacity: 358400
browser.cache.disk.smart_size_cached_value: 358400
browser.cache.disk.smart_size.first_run: false
browser.cache.disk.smart_size.use_old_max: false
browser.cache.frecency_experiment: 3
browser.fixup.domainwhitelist.monoprice: true
browser.places.smartBookmarksVersion: 7
browser.sessionstore.upgradeBackup.latestBuildID: 20141115210315
browser.startup.homepage_override.buildID: 20141115210315
browser.startup.homepage_override.mstone: 36.0a1
browser.tabs.remote.autostart.1: false
browser.zoom.siteSpecific: false
dom.mozApps.used: true
extensions.lastAppVersion: 36.0a1
gfx.canvas.azure.backends: skia
gfx.content.azure.backends: skia
layers.acceleration.force-enabled: true
layers.offmainthreadcomposition.async-animations: true
layers.offmainthreadcomposition.enabled: true
media.autoplay.enabled: false
media.gmp-gmpopenh264.enabled: false
media.gmp-gmpopenh264.lastUpdate: 1412264848
media.gmp-gmpopenh264.path: /home/kolya/.mozilla/firefox-trunk/oysdo5ii.default/gmp-gmpopenh264
media.gmp-gmpopenh264.version: 1.1
media.gmp-manager.lastCheck: 1416323038
mousewheel.withcontrolkey.action: 3
network.cookie.prefsMigrated: true
places.database.lastMaintenance: 1416290576
places.history.expiration.transient_current_max_pages: 104858
plugin.disable_full_page_plugin_for_types: application/pdf
plugin.importedState: true
plugin.soname.list:
plugin.state.flash: 1
plugin.state.libnpgoogletalk: 1
plugin.state.libnpo1d: 1
plugin.state.libpipelight-silverlight: 1
print.print_bgcolor: false
print.print_bgimages: false
print.print_colorspace: default
print.print_downloadfonts: false
print.print_duplex: 0
print.print_evenpages: true
print.print_in_color: true
print.print_margin_bottom: 0.5
print.print_margin_left: 0.5
print.print_margin_right: 0.5
print.print_margin_top: 0.5
print.print_oddpages: true
print.print_orientation: 0
print.print_page_delay: 50
print.print_paper_data: 0
print.print_paper_height: 279.40
print.print_paper_name: na_letter
print.print_paper_size_type: 1
print.print_paper_size_unit: 1
print.print_paper_width: 215.90
print.print_plex_name: default
print.print_resolution: 600
print.print_resolution_name: default
print.print_scaling: 1.00
print.print_shrink_to_fit: false
print.print_to_file: false
print.print_unwriteable_margin_bottom: 13
print.print_unwriteable_margin_left: 13
print.print_unwriteable_margin_right: 13
print.print_unwriteable_margin_top: 13
privacy.sanitize.migrateFx3Prefs: true
security.disable_button.openCertManager: false
security.disable_button.openDeviceManager: false
storage.vacuum.last.index: 1
storage.vacuum.last.places.sqlite: 1413817069

Important Locked Preferences
----------------------------

JavaScript
----------

Incremental GC: true

Accessibility
-------------

Activated: false
Prevent Accessibility: 0

Library Versions
----------------

NSPR
Expected minimum version: 4.10.7
Version in use: 4.10.7

NSS
Expected minimum version: 3.18 Basic ECC Beta
Version in use: 3.18 Basic ECC Beta

NSSSMIME
Expected minimum version: 3.18 Basic ECC Beta
Version in use: 3.18 Basic ECC Beta

NSSSSL
Expected minimum version: 3.18 Basic ECC Beta
Version in use: 3.18 Basic ECC Beta

NSSUTIL
Expected minimum version: 3.18 Beta
Version in use: 3.18 Beta

Experimental Features
---------------------

Sandbox
-------

Seccomp-BPF (System Call Filtering): true
Media Plugin Sandboxing: true


Please let me know if I can provide any more info. I'm glad to help!

Comment 7

[Chris Peterson [:cpeterson]]

Can you try disabling Adblock Plus or not using CTRL+PGUP? We know that Adblock Plus has caused some e10s performance problems before.

Comment 8

[Nikolay Martynov]

Well, I do not think I can stop using ctrl+pgup. I mean the point to to load those tabs so I could switch between them quickly. I also have several tab grous that sit on background that do not get loaded usually, so disabling restore_on_demand will degrage my experience.

Also even without ctrl-pgup e10s seems to be slower than non-e10s - see comment #4 above.

I'll try disabling adblock to see if it helps and post results.

Comment 9

[Nikolay Martynov]

So, I've tried disabling add block and it doesn't seem to make much of a difference:
  - Loading is still (subjectively) slower than without e10s.
  - Only one active tab out of several opened windows gets loaded, others get 'spinner' forever until manually reloaded.
  -- And switching between tabs when other tabs are loading is very slow.

The easier way to illustrate the latter point is as following. Assume you have two windows with several tabs on them. After you start firefox you load all tabs in window A, you wait for them to finish loading. Now when all tabs are loaded in window A you go into window B and go though all tabs there - they start loading. Then you switch back to window A and try switching tabs there. Even though all tabs have finished loading in window A before switching tabs there now is very slow and I see 'spinner' for a very prolonged amount of time. Without e10s UI might be laggy but I do not get spinner and I switch to other tabs almost instantaneously.

Comment 10

[Bill McCloskey [inactive unless it's an emergency] (:billm)]

If you disable all your add-ons, does this still happen?

Comment 11

[Nikolay Martynov]

Yes, it still happens. When I disable all addons:

Completely loading tabs in one window and then starting loading tabs another window and switching tabs in first window still shows 'spinner' in first window for tabs that have been loaded. Maybe spinner is shown for a bit less amount of time though.

If I have two windows one of them shows 'spinner' forever after FF start up.

Comment 12

[Tracy Walker [:tracy]]

Nikolay,  much work has gone into e10s since your last report here.  Is this bug still a notable issue for you?

Comment 13

[Nikolay Martynov]

Yes, it seems to be much better now.

Multiprocess start up with many tabs is still not fast, but it is almost as fast as with single process.

Thanks!

Comment 14

[Guilherme Lima]

Just my 2 cents: I think the load of the content process is what makes things so much slower.

To reproduce this, just close all tabs anytime but leave a about:memory tab open.
The paste on the url bar: https://bugzilla.mozilla.org/show_bug.cgi?id=1096756 and hit enter.
It will be a few seconds until something loads.

For me, it took 6 seconds from hitting enter and the tab show a sign of loading content of the page. (Win 7 - 3 years old Intel i3 notebook).

Comment 15

[Brad Lassey [:blassey] (use needinfo?)]

fixed according to comment 13