Firefox 3 Performance Gets a Boost

jason writes "Mozilla has been working hard at making Firefox 3 faster than its predecessor, and it looks like they might be succeeding. They've recently added some significant JavaScript performance improvements that beat out all of the competition, including Opera 9.5 Beta. And it comes out to be about ten times faster than Internet Explorer 7! Things are really starting to fall into place for Firefox 3 Beta 4 which should be available in the next week or two."

Safari

[adam1234 (696497)]

Hopefully it can best Safari's Javascript performance. Firefox is pitifully slow compared to WebCore's javascript core.

Re:Safari

[CRCulver (715279)]

Surf with NoScript and the JavaScript engine will be invoked much less than usual.

Re:Safari

[0xygen (595606)]

But with NoScript it's not so much surfing as paddling through the mud!

Re:Safari

[utopianfiat (774016)]

lynx = floaties and a shark-filled lagoon?

Re:Safari

[prestomation (583502)]

According to TFA, Safari is beat out by Firefox 3 beta 3 and 4, and Opera.

Re:Safari

[by Anonymous Coward]

Would it be fair to say the following actually occured to obtain these javascript performance improvements:

1. instrumented firefox (PGO technology)
2. ran the stinking benchmark with the instrumented code
3. used the feedback from the benchmark to automatically compile an optimized version of firefox optimized specifically for the benchmark.

4. Publish results of said benchmark for all to oooh and awwww over.

Isn't this as pathetic and useless as vendors manually tweaking their 3D drivers to artifically raise performance figures displayed in 3dmark? Did I totally misread TFA?

Re:Safari

[luserSPAZ (104081)]

No, we're just profiling on browser startup/shutdown right now. I did do a build profiled on the benchmark, and it was pretty fast, but that's probably overkill. Mostly we just want to hit enough common code paths to make things faster. Turns out sunspider perf correlates pretty nicely to overall JS speed, since the benchmark is made up of real world code that people complained was slow.

Re:Safari

[Idiomatick (976696)]

I know, it really sucks that people don't write books anymore since the printing press came out. The whole industry collapsed, you probably haven't even heard of these jobs:

* Copyists, who dealt with basic production and correspondence
* Calligraphers, who dealt in fine book production
* Correctors, who collated and compared a finished book with the manuscript from which it had been produced
* Rubricators, who painted in the red letters
* Illuminators, who painted illustrations

Ohhhh wait, people still write books and the industry didn't collapse. It just changed. I'm sure in 50years we'll be saying 'wtf was a publisher again?'. And nothing of value will be lost. Artists have the HUGE opportunity of being able to cut out the middle men (there are lots of them) with current technology. With less hands in their pockets they will make big money from live shows and bigger profit from merchandise as well as profits from ad supported downloads and site page views. Artists will NOT starve, i don't see how cutting away the massive corporations which artists are carrying on their back atm will hurt the artists.

Re:Safari

[A beautiful mind (821714)]

RTFA (or just glance below):

1. Firefox 3 Nightly (PGO Optimized): 7263.8ms
2. Firefox 3 Nightly (02/25/2008 build): 8219.4ms
3. Opera 9.5.9807 Beta: 10824.0ms
4. Firefox 3 Beta 3: 16080.6ms
5. Safari 3.0.4 Beta: 18012.6ms
6. Firefox 2.0.0.12: 29376.4ms
7. Internet Explorer 7: 72375.0ms

The results are generated by using the Sunspider JS benchmark suite.

Re:Safari

[GenKreton (884088)]

These benchmarks are definitely lal done on a windows box, because if you compare the performance of JS in Firefox on Linux and Windows it is like night and day... I don't know why JS on Linux needs to be so much worse.

Firefox 3 beta 3 on Linux is great

[LingNoi (1066278)]

I had these problems with Firefox 2 exactly running the default Ubuntu version. I downloaded the firefox 3 beta 3 from the website and it is so much better it is unbelievable.

- Where as before FF2 would use around 500 meg it now only uses around 50 meg
- Flash no longer crashes the browser
- Javascript no longer crashes the browser
- Those long pauses as it is doing something that stalls the browsers operation are gone.

I couldn't believe the difference.

Re:Firefox 3 beta 3 on Linux is great

[Curtman (556920)]

I would add to that list:

- Ctrl-MouseWheel zooming scales the images as well so the pages look normal without text overlapping the graphics

That feature alone is worth upgrading for.

Re:Safari

[dgatwood (11270)]

How about testing with a WebKit nightly?

Why is this marked as troll?

[Overly Critical Guy (663429)]

Why is the parent comment marked as troll? It was reported a few weeks ago that the next version of Safari, 3.1, would see major JavaScript performance gains due to the latest WebKit builds. This article uses the beta Windows 3.0 version to compare to.

Re:Safari

[by Anonymous Coward]

The Webkit nightly builds are significantly faster. I don't have the same machine they've tested on obviously, but for comparison purposes here's the current release vs. the most recent nightly build on my Mac OS X 10.5.2 machine:

Safari 3.0.4: 10758.4ms +/- 0.5%
WebKi r30628: 3390.0ms +/- 0.3%

If the performance gain percentage is comparable on their test machine (big if, granted) the comparable time would be 5675.8 ms, 22% faster than the PGO Firefox build.

Re:Safari

[TheNetAvenger (624455)]

(or just glance below):

1. Firefox 3 Nightly (PGO Optimized): 7263.8ms
2. Firefox 3 Nightly (02/25/2008 build): 8219.4ms
3. Opera 9.5.9807 Beta: 10824.0ms
4. Firefox 3 Beta 3: 16080.6ms
5. Safari 3.0.4 Beta: 18012.6ms
6. Firefox 2.0.0.12: 29376.4ms
7. Internet Explorer 7: 72375.0ms
The results are generated by using the Sunspider JS benchmark suite.


This looks great, but everyone should notice a couple of things that may not be obvious.

1) Sunspider JS benchmark is designed by Apple developers and they use it to show the world how much faster Safari is, however Opera seems to outpace the Safari developers even with their own tests. However, yes some of the benchmarks used are 'picked' to favor Safari, and some are 'extended' to hurt IE.

2) Sunspider over does the tests of the Append String performance problem to make IE look worse than it really is. IE's JScript is coded as JScript was designed, and because of this, it doesn't optimize string append operations by using newer code. So by using this text extra, it artificially make IE look horribly slow. IE8 and possible additional IE7 releases are spending time optimizing the base JSCript code from the original implementations/specifications.

http://blogs.msdn.com/jscript/archive/2007/10/17/performance-issues-with-string-concatenation-in-jscript.aspx [msdn.com]

3) If you remove the 'string' routine from the test, IE7, consistently outperforms Firefox 2.0, and is very close to even Safari for with the results were cherry picked.

http://www.codinghorror.com/blog/archives/001023.html [codinghorror.com]

4) Some of the numbers are quite questionable as to the validity. For example IE7 is given 72375 in this article, and yet the slowest machine our tech lab has ever benchmarked is 2x the speed, and this is on a very old AMD 1ghz machine that barely runs Vista in which the test yeilded the horrible results. So where did they get the 72375 number from? A Pentium 200?

Again reference this link so see that even this person's results are no where near the 75K ms time reported for IE.

http://www.codinghorror.com/blog/archives/001023.html [codinghorror.com]

So it is quite questionable and inaccurate to try to portray IE7 as 10x slower, when without the 'emphasized' string append slowdown in IE7, it is faster than FireFox 2.0 and within a few 'ms' of even Safari and the new FireFox 3.0 results.

Good job to the FireFox team, btw.. Also does anyone have benchmarks of the new FireFox using a non-Apple test suite?

Re:Safari

[A beautiful mind (821714)]

I don't get it. They are doing exactly what I wanted about Firefox in 3.x.

While new features can be nice, I couldn't name a feasible feature that a significant number of people would want and it's not in core Firefox or in an extension already. What I want from Firefox now is to provide the existing features in a secure, stable, fast and memory conserving way, in this order. Heck, I've turned off most of the new features in Firefox 2.x and wished they'd fix some annoying bugs instead. In 3.x the developers did a lot of work to remedy a lot of those bugs and issues, so big big kudos for them!

Cleaner code matters - it results in less bugs and security vulnerabilities, easier to add features and most likely better code.

Re:Safari

[CajunArson (465943)]

Yeah great, was anyone complaining about the speed?

Actually many people (myself included) were complaining about speed, and in some cases new "features" are just bloat. One feature that I would LOVE to see is to have isolation between tabs so that if one page in one tab causes a crash, the other tabs would be unaffected and the browser could continue. A multi-process model with better isolation could do this, and would also make more efficient use of multi-core systems (since FF is notoriously single-threaded, have a single thread per-tab instead of per-browser). FF does crash, and while sometimes a third party plugin is to blame, I really don't care about pointing fingers just in getting the browser more reliable.

Re:Safari

[Mr. Spontaneous (784926)]

They've been throwing around the idea of multithreading for Firefox 4, but right now its still in contention, I think, because it has to be done right. I recall reading some dev blogs that said they'd jump ship if the team decided to expose the threads to extension developers.

Re:Safari

[steelfood (895457)]

Threads add a whole new dimension of complexity to the engine. The "right" way to do it may not even exist.

Extensions definitely should not have direct access to the threads. It would be an absolutely terrible idea. In fact, extensions shouldn't even know that there's multithreading going on behind the scenes. At best, extensions would be able to indirectly spawn threads and manipulate the spawned threads in a roundabout manner through that context using a thread-safe API.

Re:Safari

[AaronW (33736)]

Just because I have a couple gigs of memory doesn't mean I want Firefox to consume it all. I run more than just a web browser. I don't want a lot of caching anyway, since most pages I hit are dynamic and I don't use back very frequently. I don't want a program to look and see I have a couple gigs of memory and assume it can use it all.

-Aaron

Re:Safari

[squidinkcalligraphy (558677)]

As has been often discussed, Linux's reporting of memory usage is inaccurate. Firefox uses threads; in Linux each thread is treated as a process, and appears as such to programs like top, appearing to use it's own memory allocation plus the shared memory common to all firefox threads. So it _looks_ (approximately) like the memory being used all up is (thread's mem + shared mem) * Number of threads, where in fact it is thread's mem * number of threads + shared mem.

Re:Safari

[dwater (72834)]

...I suspect our brain tries to do shortcuts...

What? Are we 'the borg' now? ...or some kind of massive Siamese poly-tuplet.

Re:Safari

[billcopc (196330)]

Are you sure it's actually swapping to disk, or is it merely reserving swap space in the map ?

Example: I have 8 gb in this system. Right now I only have FF and Thunderbird running (+ a few background processes). Current "commit charge" is 475mb, of which Firefox is using 150mb. The system says I'm using 280mb of swap, but it's not actually thrashing the swap disk at all. That swap space is reserved, presumably because it represents 280mb of idle memory that is eligible to swap out, should another process need it.

Windows allocates virtual memory quite aggressively (when properly coded). If a process requests 500mb, but only really uses 100mb of it, the remaining 400mb will be "allocated" to swap while the real memory remains available to other processes. The moment a memory page is accessed, it is marked "dirty" and moved to real memory.

It's very much like sparse files, where unused or 0-filled pages don't take up any physical space (except for the map entry). That's how virtual memory is supposed to work, and it lets developers simplify their code by not having to worry too much about the physical arrangement of memory. It's also partly why you should never run a system without a swap file, even if it has tons of memory. I've probably never used all 8 gb in my system, but I still keep a (small) swap file. If I didn't, and that process allocates 500mb, Windows needs to dedicate 500mb whether or not it is actually in use. It reminds me of real-estate players, who can "buy" million-dollar buildings with a relatively small amount of capital, the rest on credit. Swap is like a line of credit for the OS.

Linux probably does the same thing, but I'm not as knowledgeable about its inner workings.

Re:Safari

[99BottlesOfBeerInMyF (813746)]

They should have run it against the latest webkit.. it's supposed to be pretty fast..

Okay, I ran it on OS X anyway. I'm too lazy to run it on Windows too :) Here are the results [slashdot.org]. The new version of Webkit/Safari does beat the nightly of Firefox, but it is close and they're both a lot better than any regular release.

I tried Firefox 3 today

[celardore (844933)]

To be honest, I hate it. WTF have they done with my handy URL bar? It used to be a place where I could type "slas" and get the slashdot URL come up. Even worse for "news", as it "handily suggests" all the pages in my history that have "slas" or "news" in my history.

Heads up for all those trying Firefox 3 is Oldbar [google.co.uk]. I suggest you get it if you don't like the new 'innovations' by Mozilla Corp.

Re:I tried Firefox 3 today

[christopherfinke (608750)]

It learns as you use it. Type 'slas' and choose Slashdot from the list. After doing that once or twice, Slashdot should automatically float to the top each time after that.

Re:I tried Firefox 3 today

[arth1 (260657)]

Sorry, copying the bad eyecandy from Microsoft doesn't classify as an improvement in my opinion.

The learn-as-you-go menu behaviour which they copied from Windows didn't work well in Windows either. The main problem is that it causes inconsistent behaviour. Repeating something doesn't necessarily give you the same menu items. It's good for newbies who read every single line before choosing one of them, but it's very bad for people who memorize what they do so they can repeat it quickly without even looking.

Re:I tried Firefox 3 today

[TheSkyIsPurple (901118)]

>I don't want software that tries to think for me.

In which case you don't want the browser to autocomplete the URL for you at all, and the fact that it finds seemingly irrelevant matches shouldn't matter.

Re:I tried Firefox 3 today

[by Anonymous Coward]

Funny, I have been using it for a couple months now (nightlies) and I absolutely love the awesomebar. Just typing "s" gets me slashdot. the various environments I work with can be gotten with "l" (localhost), "d" (the development server), "bug" "sprint" "-1h" "me" (our bug tracker), "qa" (our qa environment), etc.

Best of all, if I visit any site and then want to get back to that site again sometime, all I need to remember is something in the title or url of the page I was at.

Re:I tried Firefox 3 today

[balster neb (645686)]

I agree with you totally. I've been toying around with the Firefox 3 Betas for a couple of weeks and I think the awesomebar is the best new feature. It's not broken -- it's different. Once you get used to it, finding pages you frequently visit becomes much easier.

Say you visited the Wikipedia page on the Tunguska event a couple of weeks. If you want to revisit the page, all you do is start typing the first few letters of "Tunguska" and the page comes to the top of the list. With the old type of address bar, you'll have to type the whole Wikipedia URL or search your browser history separately. This speedup is well worth the relatively shallow learning curve.

I find it pretty stupid to compare this feature with Windows' "adaptive menu" feature. There's only a superficial resemblance. Remember, the traditional address bar still "learns" in the way you hate by ordering URLs by the frequency with which you visit them. What is it with the Slashdot crowd and being insanely conservative about their software?

Re:I tried Firefox 3 today

[caspy7 (117545)]

You've only tried it for one day and you hate it?
I think I understand.
You see, the new location bar learns. Though this silly new 'innovation' does indeed search through the URLs *and* titles of bookmarks and history, it also learns what you select the most. Give it a few more days and slashdot should come to the top of the list.
I experienced the same thing in the beginning.

When I bookmark page now I try to throw on a couple common sense tags that way when I type the tag in the location bar in the future, those bookmarks come out on top.

If you're *really* dead set on the shortest route:
1) Click Bookmarks -> Show All Bookmarks
2) Find the slashdot bookmark and select it
4) Click "More" under properties
5) Make the keyword /.
6) Close the window

Now type /. in the location bar and vwala!

Re:I tried Firefox 3 today

[upside (574799)]

You don't realize "vwala" is the de-frenchified, freedom-loving, non-retreating version of the word from the Freedom Fries crowd. It means "Behold, believe and profess. Else you will be classified an enemy combatant".

Expect to find it in a GWB-approved dictionary any day now.

Firefox 3 also supports new Java plug-in

[kaos07 (1113443)]

From the Mozilla blog:

Firefox 3 is going to include support for the new Java SE 6 runtime environment.

This is a new implementation of the Java Plug-In that features increased reliability, ability to specify large heap sizes, ability to select a specific JRE version to execute a particular applet, and support for signed applets on Windows Vista.

The New Plug-in is designed to work with: - Internet Explorer 6 and 7 on Windows XP and Windows Vista - Firefox 3 on Windows XP, Windows Vista, Solaris and Linux

Personally, I've been wanting to use the Firefox 3 beta for some time, primarily because of the performance and speed boosts over Firefox 2, but my favourite add-ons still aren't compatible.

Note: The new Plug-in does not work with Firefox 2, and no support is planned for this browser with the New Plug-in.

http://gemal.dk/blog/2008/02/24/firefox_3_gets_a_new_java_plugin/?from=rss-category/ [gemal.dk]

Microsoft's Biggest Mistake

[TheNarrator (200498)]

Microsoft's biggest mistake was thinking people wouldn't write complicated apps in Javascript. They supported it, in their usual half broken style, but it created the only widely deployed cross-platform system for running code that Microsoft has ever implemented. Now, with Firefox 3 running so fast javascript might become THE platform. It's hilarious because Javascript started out as such a kludgy platform and now it is becoming a serious contender if only because it's the only cross-platform thing Microsoft ever supported.

Re:Microsoft's Biggest Mistake

[chelsel (1140907)]

I have nightmare's about JavaScript being the one language to rule them all... please, let's have no such talk.

OS X Results - Spoiler Safari Wins

[99BottlesOfBeerInMyF (813746)]

Well someone had to, so I ran the numbers for OS X. All of the below were on OS X 10.5.2 running on a MacBook:

• Safari 3.0.4 - 11112.0ms
• Safari with Nightly Webkit r30628 - 3525.8ms
• Firefox Nightly3.0beta4pre - 4330.2ms
• Opera 9.26.3727 - failed (but all those that ran were slower than Safari 3.0.4 so it is the slowest overall for what worked.)

I guess if you're a Safari or Firefox person you can look forward to some really fast Javascript performance either way.

Have they discovered threads yet?

[RelliK (4466)]

Or does a single tab still cause the entire browser to freeze up?

Re:Have they discovered threads yet?

[Wolfier (94144)]

Exactly what I think.

Who cares about Javascript performance when a single script running at any speed can freeze the entire browser?

Or a few Youtube tabs can slow the browser to a halt? (Hint: Firefox REALLY need to delegate Flash rendering to an external process, something I can renice 19. Just like how Konqueror uses nspluginviewer)

Re:Have they discovered threads yet?

[multi io (640409)]

If you can find a way to magically thread javascript in a way that allows multiple windows and tabs to communicate with each other (as the DOM requires), I'm sure the mozilla folks would absolutely *love* to hear about it.

How about just implementing it? No magic needed. If the whole UI is slow and tends to lock up because it uses only a single thread, and the reason for that is that the language/runtime the UI is written in doesn't support threads, then you have three options:

• keep everything as it is, maybe pretend the problem doesn't exist
• rewrite in a language that does support threading
• extend Javascript resp. its runtime/libraries to support threading

The last one is probably the best option if you want to solve the problem, minimize the amount of work required to do so, and don't want to force all the plugin writers to use another language.

Focusing on the wrong aspects

[abigsmurf (919188)]

I always see these benchmarks and wonder "why does this matter?". The only time I ever see Javascript run too slow or tax my CPU is when it's buggy and then it'll probably throw up all sorts of warnings anyway. This is on any browser I've used and any system.

What matters to me is the imperfect implementation of Flash (it's not really their responsibility but it is their problem) which often eats up 100% CPU from random flash objects or causes firefox to freeze. Another annoyance is Firefox being frankly poor at displaying large HTML files (when you go on websites with insanely large lists for instance). Where as IE and Opera display these as the page is downloaded. Firefox, for me, freezes, much like notepad will when you open a 2meg+ file . Sometimes it'll recover and display the page after a minute or so, sometimes I have to ctrl+alt+delete.

Re:Firefox Performance

[CRCulver (715279)]

While proposed jokingly before, why not use something like PDF or flash for a fully graphical web?

Flash has more and more accessibility support, but PDF is the Page Description Format. It's meant for print output and says nothing about the meaning of the contents of the document, just how they are supposed to look on the screen and on the page.

I think that is something that could be worked on, by providing an open standard for the files that can be parsed easier than html.

The good thing about tag-based formats like HTML is that--provided someone's following the standard--they can be fairly easily parsed regardless of the output format. With XHTML, you can read stuff on your screen, the blind can use screen readers, and web developers can easily extract and transform elements from a given document things are good as they are.

Finally, why do you think PDF = lean and mean? Acrobat proves that a PDF reader can get hideously bloated.

Re:Firefox Performance

[Vombatus (777631)]

but PDF is the Page Description Format.

I could have sworn that PDF was Portable Document Format. All your other points about it are correct though.

Re:JavaScript, huh.

[jesser (77961)]

The benchmark used in this article is a JavaScript benchmark, but PGO was enabled for most components of Firefox, not just the JavaScript engine. And even if only the JavaScript engine improved in speed, you'd see a speed boost despite having JavaScript disabled in web pages, since parts of Firefox itself are implemented in JavaScript.

Re:Memory leak?

[bunratty (545641)]

No, recently the developers have found that few leaks are left, so that to reduce memory usage further they had to change their focus to reducing fragmentation. Originally, the problem was leaks, it's just that once the worst ones were fixed fragmentation became responsible for a larger fraction of the memory usage. This a continuation of people trying to find one single cause of high memory use. As I and others have been saying for years, there is no one cause. There is no "the memory leak" or "the memory issue", just as there is no "the crash problem" or "the security problem".

Re:Memory leak?

[fontkick (788075)]

There is no "the memory leak" or "the memory issue", just as there is no "the crash problem" or "the security problem"

Once upon a time there was this OS named Windows Millennium Edition, also known as "the" in your examples above.

Re:OSX?

[BrainInAJar (584756)]

I have. Still ass. doesn't go lighter when it's backgrounded, stays the same dark grey as if it were foregrounded.

Open-Source seems good for getting a job 90% finished and completely ignoring the 10% polish required to make it an app of the same quality as closed-source

Re:IE7 is just slow anyway

[A beautiful mind (821714)]

3* tabs should be enough for everybody.

*For older versions of IE, 3 equals 1.

Re:About time

[LingNoi (1066278)]

You may freely use and distribute the code presented in these tutorials under any license EXCEPT the GPL or any other license which denies authors their right to do as they please with their own code.

Hypocrite

Re:stalling

[roca (43122)]

Actually in Firefox 3/Gecko 1.9, external CSS loads do not block the parser. Woohoo! However, we do block the parser if the page tries to execute script while there are pending CSS loads.