Guides To Analyzing WebKit Performance

– Looking at the internals –

Holger Freyther

Developer, WebKit Project

Jim Huang ( 黃敬群 ) <jserv@0xlab.org>

Developer & Co-founder, 0xlab

April 28, 2011 / Android System Development Forum

Rights to copy

Attribution – ShareAlike 3.0You are free

to copy, distribute, display, and perform the workto make derivative worksto make commercial use of the work

Under the following conditionsAttribution. You must give the original author credit.Share Alike. If you alter, transform, or build upon this work, you may distribute the resulting work only under a license identical to this one.

For any reuse or distribution, you must make clear to others the license terms of this work.Any of these conditions can be waived if you get permission from the copyright holder.

Your fair use and other rights are in no way affected by the above.License text: http://creativecommons.org/licenses/by-sa/3.0/legalcode

© Copyright 2011 0xlabhttp://0xlab.org/

contact@0xlab.org

Corrections, suggestions, contributions and translations are welcome!

Latest update:April 28, 2011

• Optimize WebKit for the Content?

• Optimize the Content for WebKit?

The Goal of This Talk(1)

• For optimizing content, see the remote inspector work• Another example: Opera Mobile Accelerator

http://www.opera.com/press/releases/2004/06/09/• This talk will be about approaching WebKit

The Goal of This Talk(2)

• Take the fear from working on a big project.

• Show ways to approach the codebase.• Establish ''do not guess but measure''

attitude.

The Goal of This Talk(3)

Agenda (1) What is WebKit (project)?

(2) Android & WebKit

(3) How to prepare Android to measure?

What Groups are Responsible for Managing it?

8

GraphicsNetworking

WebCore

WebKit GTK WebKit Mac QtWebKit

JavascriptCore

WebKit Architecture

9

cairo & pangocurl / soup

WebCore

WebKit GTK

JavascriptCore

(Flexible) WebKit Implementations

Gtk+ windowGtk+ windowGtk+

QtWebKit

WebCore event Refresh the surface(expose event)

cairocairo„expose event“ callback

Gtk+ surfaceGtk+ surface

WebKit drawn to Gtk+

Gtk+ program

Gtk+

QtWebKit

WebCore event

Refreshing the surface

cairo

„update-requested“ callback

cairo surface (every supported cairo surface)

WebKit drawn to Gtk+ applications

„update-requested“ event

• ~80 Reviewers• ~110 Committers• 8 ports in the tree• Apple and Google (Chromium) are major

contributors• Many commits per day

What Is the WebKit Project?

• Check http://webkit.org/projects/goals.html• Goals vs. Non-Goals

• Content Engine, Security, Performance and more• Every change needs review, no performance

regression allowed• But performance tests are private due to

copyright laws

How Does WebKit Work?

• Android is not involved with the WebKit project• Android style open source model

• Android is using Chromium as upstream• Android does not include the data for Quality

Assurance (tests)• Who is fixing known security issues in the Android

code?

Android & WebKit

ChromiumAndroid'sWebKit fork

WebKit

JNIJNI

Skia bridge

Gtk+

WebKit

WebCore event Refresh the surface(expose event)

skiaskia

SurfaceSurface

WebKit in Android

Android.webkit.WebViewCoreandroid.webkit.WebView

...

v8

libpixelflinger

SurfaceFlinger

libui

SurfaceOverlay Camera EglWindowsformat Key / Event

OpenGL|ES

android. graphics. Canvasandroid.view. Surface

Graphic JNISurface JNI

android.view. View

GLSurfaceView javax .microedition .khronos .openglesUI Element

OpenGL JNI

com.google.android.gles_jni

FrameBuffer Driver

NativeFramework

JAVAFramework

Event Input Driver

skia

libui +SurfaceFlinger +

libpixelflinger

libui +SurfaceFlinger +

libpixelflinger

View and Canvas → SurfaceView and Canvas → Surface

Java Graphicsis implemented

via JNI

Java Graphicsis implemented

via JNI

• What is performance?• How to measure it on GNU/Linux?

• How to do it on Android/ARM?

Moving To Performance Now

• Amount of useful work accomplished• Examples:

• how fast does the page load?• How many frames per second are drawn?• How little/much bandwidth is used?

• Optimization mostly trade off between Memory and CPU usage

Computer Performance

• Do not assume, meassure it!• Have a manual or automatic testcase• Observe the system while running the testcase• Analyze the situation, make changes• Repeat until considered good enough

Performance Experiments

• Easy to setup• Open a site and wait, or scroll• Good for getting an idea• Bad for repeating and comparing results

Performance Experiments – Manual

• More difficult to create• Requires stable content• Should allow to compare results• Talos and others as a framework

• https://wiki.mozilla.org/Buildbot/Talos

Performance Experiments – Automatic

• perf (new way)• oprofile (old way)• Both are sampling profilers

• ARM Performance Counter; PMU (Performance Measurement Unit)

How to Observe on GNU/Linux

ProfilingAnalysisEngine

Prof.out

Profiling data preparation

Trigger Profiling

Profiling

Simulate

Profiling

Qualified?

END

Tune Performance by CPU Configuration

Yes

No

Evaluating and Tuning

# prepare the setup

$ rm -rf /var/lib/oprofile

$ opcontrol --start-daemon -p library -c 10

# run the app once to force loading it from nfs into the cache

$ ./tst_something

# start profiling

$ opcontrol –start

$ ./tst_something -iterations enough

# stop profiling

$ opcontrol -h

# generate reports

$ opreport -l

$ opreport -c …

Reference oprofile usage

Profile Result (A)

Profile Result (B)

Profiler sample

http://anton.ozlabs.org/blog/2009/09/04/using-performance-counters-for-linux/http://anton.ozlabs.org/blog/2009/09/04/using-performance-counters-for-linux/

perf is powerful.# perf record -s CMD Error: perfcounter syscall returned with -1 (Function not implemented) Fatal: No CONFIG_PERF_EVENTS=y kernel support configured?

perf is powerful.# perf record -s CMD Error: perfcounter syscall returned with -1 (Function not implemented) Fatal: No CONFIG_PERF_EVENTS=y kernel support configured?

• for Native libraries →• Use 'perf' built without libperl, libpython• oprofiled and opcontrol are there, CPU data is

missing• Binaries for ARM need frame pointers to have

backtraces• Java part is the performance hell always.

• traceview is a great tool for Java performance analysis.

• JVMTI / JDWP (Java Debug Wire Protocol, normally spoken between a VM and a debugger)

How to Do On Android/ARM?

• Upload some more files• Start oprofile with opcontrol on the device• Run the test on the device• Analyze with opreport on the PC

How to Do On Android/ARM?

# Overhead Command Shared Object Symbol# ........ ............... ..................... ......# 89.23% system_server 2b0c6c [.] 0x000000002b0c6c 1.26% MLVdo_thread [kernel_helper] [k] 0x0000000017aa90 1.05% d.process.acore libskia.so [.] S32A_Opaque_BlitRow32_arm 0.83% d.process.acore libcutils.so [.] android_memset32 0.63% system_server libc.so [.] memcpy 0.63% d.process.acore libc.so [.] memset

system_server is the process name of Android Framework runtime. It occupies most of CPU resources, but it is hard to figure out details only by native tools like perf.

system_server is the process name of Android Framework runtime. It occupies most of CPU resources, but it is hard to figure out details only by native tools like perf.

We can always optimize known performance hotspot routines such as S32A_Opaque_BlitRow32_arm but should be measured in advance.

We can always optimize known performance hotspot routines such as S32A_Opaque_BlitRow32_arm but should be measured in advance.

• What to measure? Loading, Painting, Scrolling?• No excellent benchmark suite available due to

copyright issues• Some frameworks are available, but mostly

manual work

Picking or Creating a Testcase

• Small Framework from the University of Szeged• Can load pages and count the time• Provides a summary with error interval

Introducing The Methanol Framework

http://webkit.sed.hu/blog/20101216/benchmarking-qtwebkit-v8-linux

• Webpages need to be converted• Everything loaded from the same URL• http://gitorious.org/methanol

Issues With Methanol Framework

• Using methanol with the example page• Executing it in the Browser• Observing it with oprofile/perf

Putting Everything Together

http://0xlab.org