449919 - Major rendering performance regression vs FF2, specially when scrolling page content

Status: UNCONFIRMED

(Core :: Web Painting, defect)

Description

[A. Glez. B.]

User-Agent:       Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.0.1) Gecko/2008070208 Firefox/3.0.1
Build Identifier: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.0.1) Gecko/2008070208 Firefox/3.0.1

Considering CPU load and application responsiveness when performing rendering operations of both UI and Content, FF3 has suffered a mayor performance regression compared to FF2. The high CPU load is observed only when FF3 is visible, never when in background or iconified. The lack of responsiveness can be best seen when scrolling page content or switching open menus by moving the mouse pointer. This issue is FF3 specific since FF2 performs just like any other native application.

Reproducible: Always

Steps to Reproduce:
1. Close all running applications and stop all background services.
2. Open FF3 with a clean profile.
*3. Open a menu (like File) and move the mouse pointer over the others.
4. Disable smooth scrolling and reset the zoom.
*5. Open the given URL or a local copy from your HDD.
*6. Scroll the contents.
Actual Results:  
*3. The FF3 UI responsiveness is poor and the CPU load is high. Opening a menu or switching open menus shows a considerable delay.
*5. High CPU load while the page is loading. Disabling the "Always show the tabbar" helps. 
*6. When scrolling the content of the page is garbled and zooming or enabling smooth scrolling makes it worse. 

Expected Results:  
Low CPU load, normal menu switching and scrolling: just like FF2.

I noticed the issue on the very FF3 Launch Day but on my hardware (see below) was almost tolerable because more content/functionality means more overhead. That was until yesterdey I tried to scroll back and forward the XML-Schema examples and the delay was unbearable, not just annoying because it's just text with a straightforward layout and no images nor animations.

I tried a clean profile and safe-mode to determine if caused by something else but the result was the same. So I asked myself if it actually was a hardware issue so I downloaded and installed (again) FF2 with a clean profile and the performace is fine: no noticeable delay, no high CPU load, perfect scrolling.

Please note the Trident Blade 3D is the WORST video card ever made into a P3 board and since I'm used to it (I've had it for quite some time now) it's not unusual for an issue like this to go unnoticed for me and for people with better hardware who won't notice this because it's compensated with high CPU/VIDEO power.

The thing is FF2 performance is all right so why not FF3's? Right now I'm wondering how to bring the bookmarking system and the awsombar back to FF2 until this is fixed.

Finaly there are about 20 other open or unconfirmed bugs about performance issues with FF3 on different platforms (keywords: performance, cpu load, delay) wich are or might be closely related to this.

Hardware:
CPU: Intel Pentium III EB 1GHz (Coppermine)
RAM: 1GB (two 512MB 133MHz Kingston PartNo: 64MX64PC133CL3168)
Motherboard: AOpen MX36LE-UN
Chipset: VIA Apollo PLE133 (VT8601) rev. 05. Using HyperionPro v5.12a drivers.
Video: Trident Blade 3D (Onboard. AGP 4x 8MB VRAM) using XP drivers.

Software:
OS: Microsoft Windows XP Version 2002 ServicePack 2
FF2: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.8.1.16) Gecko/20080702 Firefox/2.0.0.16
FF3: Mozilla/5.0 (Windows; U; Windows NT 5.1; en-US; rv:1.9.0.1) Gecko/2008070208 Firefox/3.0.1

Comment 1

[A. Glez. B.]

Additional info:
If leave only the first 200 lines of the page FF scrolls better so this raises a question.

¿Does FF renders the hole page or just the visible part?
If the answer is the hole page then the issue is critical because every new feature (HTML, CSS, etc) will produce a little performance regression which will finally add up to an extremely poor overall performance.

Comment 2

[A. Glez. B.]

FF3.0.2 performs a lot better than 3.0.1, but it's not yet as fast as FF2.

Interestingly I found this:

Speed. Firefox 3 has gotten a performance boost by completely replacing the part of the software that handles drawing to your screen, as well as to how page layout work is handled.

at http://developer.mozilla.org/en/Firefox_3_for_developers#Performance. This may be the explanation for the issue.

Comment 3

[Wayne Mery (:wsmwk)]

dupe of bug 437749?  => minor based on comment 2

almost no bugs mention trident so I suspect you'd get better info by consulting a support forums. but here are the bugs:
bug 328921
bug 25054
bug 79889
bug 293611
bug 391528
bug 188853
bug 362093
bug 449919
bug 456999
bug 457365
bug 460564

Comment 4

[A. Glez. B.]

The Trident video card is just a low perf card. I fully specified the hardware to let others know why even if the issue is there it may go unnoticed on a more powerful hardware configuration.

Bug 437749 is right on the spot but repaint is not the only thing slowing down FF UI. If every time I scroll the page FF needs to repaint/layout the hole page instead of just the visible part, that is a bug. I'm just describing the effect produced by a series of changes gone into FF3 as I see them. I just can't or don't have the knowledge to pinpoint the specific causes.

But adding up all those NEW/UNCONFIRMED bugs and some posts on PlanetMozilla you can see a real performance regression when rendering/painting UI and pages or just google for "site:mozillazine.org slow FF3" and see.

Comment 5

[Boris Zbarsky [:bzbarsky]]

Hmm.  Those are almost the same specs as my old machine (that was a P3-733, the rest about the same including the chipset).  But I don't recall particular issues with scrolling on W3C specs...

vlad, any idea what might be going on here on Windows?