Firefox 4.0 Beta 7 - Super fluid, beats everyone on sunspider | Hacker News Reader

Firefox 4.0 Beta 7 - Super fluid, beats everyone on sunspider

mozilla.com

132 pointsashish0185 comments

Browser vendors talk about benchmarks all the time but what really matters to me is startup and closing times; here Chrome really shines and FF3.6 is slow as hell.

It's probably about perception; maybe the browser should show a window even if it's not really ready (and maybe there is a problem with plugins and add-ons that need to initialize themselves, etc.)

Anyway, how is FF4b7 doing in this regard...?

FF4b7 is still sower to startup than Chrome for me. They are right, though ... it does feel faster on webapps and rich sites than the previous beta.

True . Same with me too .

Agreed. I can also open/close tabs a lot faster in Chrome than in Firefox...

Bootup time used to matter to me but now my laptop always sleeps - and likewise, I very rarely close all my browser tabs so the startup time is not very important to me.

It's not just about startup time, scrolling and switching tab is also a lot smoother in Chromium than in Firefox (at least on linux)

Strange, how often do you open/close your browser?

I use Firefox mostly with Selenium, to run my automated tests, so I do it numerous times each day.

After a while when there are many tabs opened, I find it easier to start anew by closing the browser (in fact I kill the task).

There is also a shared computer in my home that is used by my wife and my kids; there, browsers are constantly opened and closed since for some reason, most non-tech people (that I know -- and that includes my wife) close and reopen the browser to start a new browsing session.

Whenever Firefox memory use has ballooned to over 1GB.

Which is pretty much daily.

For me it's UI responsiveness - seems I must be one of few people in the world that care about it since I've never seen any article or benchmark on that (there was a couple about startup times), so I'm not even sure if it isn't only a matter of perception, but for me the problem with Firefox is that the UI feels slow, like there is a lag between clicking on an element and an action being taken by the program. Safari is also very bad at it (in Windows), as well as IE (but to a lesser extent) - Opera and Chrome on the other side are both great. But again, it's never a topic, so I suppose there's no progress on that in FF4, or is there?

UI responsiveness never seems to get the love it deserves. That I can count the time taken to respond to right-clicking a project in Eclipse on every machine and OS I have available is shocking. How can people use this software?

I just right-clicked on a project in Navigator in Eclipse. It opened instantly. I have about 10 projects though; maybe you have way more? Not sure what the problem might be on your end, but rest assured Eclipse does not behave that slowly for those that use it.

Or, perhaps, a single project that is enormous. Eclipse has rarely felt responsive to me, either.

I can do this with a single project on Mac OS X 10.6, Ubuntu 10.10, and Windows 7, and I have been able to repeat the observable latency on every machine running Eclipse I have ever seen, including those which were said to be 'fine' by their users.

I suspect it is more a matter of personal tolerance.

Are you doing Android development? If so, Eclipse 3.6 isn't compatible with the ADT plugin. I had the same slow downs, but moved back to 3.5, and it works fine now.

http://developer.android.com/sdk/eclipse-adt.html

Actually, the Chrome team cares a lot about responsiveness and have written and even made videos on the subject. Here's an interview with Peter Kasting (UI engineer) specifically on the topic of UI responsiveness: http://www.youtube.com/watch?v=ymgXTdWWNUU&feature=playe...!

Mozilla's wiki on that: https://wiki.mozilla.org/Perceived_Performance - including ideas about program's startup time, page loading, tabs, scrolling, etc.

Chrome only shined at startup when I first installed it. Now that I've used it for a while and its browser history has been populated, its startup is now comparable to Firefox.

Chrome is crazy with history; it seems to save everything forever. I delete its history regularly; I had not noticed it had an impact on startup time, though.

There's got to be something with their history implementations. I've got about five months worth of history in Safari and it loads after reboot in a second.

Changes noticed on Mac since Beta 6:

* The stop/reload button has moved to the right end of the URL bar.

* The status bar is gone; URLs appear in ghosted text on the right side of the URL bar when you hover over a link.

* You can enable an "add-on bar", which appears to be a replacement status bar that add-ons can add icons to. None of the add-ons I have installed (about 20) are making use of it, so the transition from overlaying the status bar to overlaying the add-on bar must not be automatic.

* In the Add-ons Manager, an explicit "Remove" button has replaced the small "X" icon that used to be used for uninstalling an add-on.

* A more colorful throbber.

I think moving the status bar to the location bar was a bad idea. The area in and around the location bar is already quite cluttered so I don't see the point. In many cases the bar isn't even big enough for the entire URL of the link being hovered over, unlike when it was in the status bar.

At least it kinda looks nice.

The performance is much better. I’m getting sub-one second app launch times, page load is much faster than safari, page render is comparable.

The app is still a bit of a bad citizen in that it doesn’t have any of the native text views allowing for system spell-check or any of the system services, and the app’s cpu usage still runs at 4% with no windows open (expected behavior is that windowless apps use 0)

And to be pedantic, there’s something still Java-y about the interface. Fonts are just a little bit bigger in places, and the main bar, despite being 7px smaller than safari’s (amazing feat, firefox dev’s) still feels chunkier.

Weird things in the interface vein: The feedback button’s dropdown hangs left automatically. Proper system behavior is to drop it down to the right, and then push it left based on the monitor’s viewport. The textfield resize grab changes the mouse cursor.

Weird stuff like that. Since it’s a beta I can understand the rough edges strapped to the oversized V12 engine. I still have nightmares about trying to run Firefox 2 on Tiger in 2005 or so.

There have been significant advances to the underlying code, see this post here by Josh Aas, Gecko platform software engineer with Mozilla Corporation:

http://boomswaggerboom.wordpress.com/2010/11/10/firefox-4-fo...

Asa Dotzler says we should 'move on' to other, more relevant benchmarks, since everyone is basically fast enough on this one: http://weblogs.mozillazine.org/asa/archives/2010/10/some_sun...

I definitely agree here - newer benchmarks are much more interesting, as sunspider doesn't really tax any VMs these days.

Browsers are going mainsteam and it spells bad news for the browser vendors and good news for the users. All browers nowadays pretty much look the same, have the same performance and try desperately to distinguish themselves with surrounding features.

The basic problem is that the browser is still "trapped in a sandbox", ie they're trying to become the OS but they're still just an application running in a limited context. If all browsers looks the same and perform the same it's going to be real difficult to differentiate yourself from the competion.

It makes sense from big guys point of view. Apple, Microsoft and Google must be present in the browser space to be competitive, especially google whose strategive agenda involves moving application from the OS to the web. For Firefox and Opera things look a bit more bleak IMO.

I think browsers will be commoditized to the degree it will be very difficult to compete, it will all come down to brand awareness and surrounding features.

My hunch is that Firefox and Opera is dying a slow death and will be irrelevant within 3 years.

I don't know about your predictions, but I think that Google's Native Client for Chrome will be a significant differentiator in bridging the desktop/web app gap once it matures and gains traction, with the help of Chrome OS too. Mozilla has some ideas for this sort of thing -- Prism is a barely-alive project that makes desktop apps out of sites and gives extra privileges and desktop APIs to the JavaScript client. I wonder if they would or can just take NaCl from Google too though, or make their own compatible platform based on LLVM.

I would use 4.0 Beta full-time if Firebug and Yahoo's Delicious extensions were updated. The memory utilization is great just like Firefox 3.6. Compare sometime the memory utilized while running many tabs in Chrome vs. Firefox over a prolonged period of time.

Update: I found Firebug has an alpha release supporting Firefox 4. http://getfirebug.com/releases/firebug/1.7X/

I've found the same, on linux x64 at least. Firefox is a little more stable when you run an ungodly amount of tabs and leave it open for days on end (yes I'm in therapy but so far it's not working). That, plus its superior Vim plugin seals the deal. Looking forward to 4.0, though I'm actually glad they decided to push back the release a few months. Stability > everything, imho.

I'm using 4.0 "full-time" and it's the nightly build :). The only thing you need is Compatibility Reporter extension. It's here: https://addons.mozilla.org/en-US/firefox/addon/15003

Seems very buggy. This is what mine looks like:

http://imgur.com/64D7h

Also, here are a few other issues:

* Navigating to www.google.com gives me a 302 page before redirecting me to the localized version (seems like it does this on other sites as well)

* The AdBlock icon is missing

* No status bar means I can't access my NoScript, MultiProxy Switch, or FireBug addons easily

* Sometimes tabs just refuse to close

* It didn't seem to want to upload the above image to imgur.com. Might have been some other problem but it worked in IE8.

* The back button doesn't work (backspace does though)

* ctrl-shift-t doesn't work ("Restore recently closed tabs" is greyed out)

* When I submitted this comment I wasn't redirected to the comment page - a blank page was all I got (same thing that happened on imgur, so I'm assuming it's a related issue)

* Sometimes I get stuck in the Tab Groups window and I can't return to the main view

This is all within 10 minutes of installation, so I'm expecting to find more bugs.

The status bar has been replaced with the addon bar. If it’s not showing up for you then look under View / Toolbars / Add-on Bar. Regarding your other problems, something’s definitely wrong because I can’t reproduce any of them. Maybe try with a new profile? Check the troubleshooting section on http://www.mozilla.com/en-US/firefox/4.0b7/releasenotes/

Looks like it was because of an incompatible addon.

Edit: NoScript appears to be problematic

> The status bar has been replaced with the addon bar.

Is there some alternate way to check where a link is pointing with a mouseover?

Yep, look in the location bar. Link targets are shown on the right hand side.

Aha sure enough! That's actually quite elegant, thanks very much.

I'm curious to see a Chrome screenshot on those wm settings. Obviously you try to minimize the the os memory footprint, so most of the UI enhancements are not really targeted for you.

You got lucky. Mine turned out like this:

http://i.imgur.com/PvjWf.jpg (Sorry, I uploaded PNG, Imgur munged it.)

Had to start in Safe Mode, resetting toolbars and themes didn't fix. Resetting all User Preferences did fix the appearance, but broke 1Password. Reinstalling 1Password (which supports Firefox 4 betas) didn't fix. (Adblock+ and Xmarks worked okay.)

TimeMachine backup from 2 hours ago to the rescue: App, Profile, and Prefs, all's good.

Can anyone elaborate on the XPCOM improvements? I thought the plan was to get rid of XPCOM all together in firefox 4.

Very nice set of improvements though. Not sure about firefox 4, but firefox 3 was a memory hog.

You can read this:

https://developer.mozilla.org/en/XPCOM/XPCOM_changes_in_Geck...

Overall, the changes are great and a long-time coming. As someone who embeds Gecko as a foundational technology for our application, the changes have been pretty bumpy as of late. Next year, I'm hoping the codebase will stabilize. I'm especially looking forward to the new single-HWND rendering code: http://weblogs.mozillazine.org/roc/archives/2010/10/bye_bye_...

>... but firefox 3 was a memory hog.

I really do not find Firefox 3 to be a memory hog at all. I find it to consume much less memory than Chrome running multiple tabs over a prolonged period of time. I have caught Chrome running way out of control with memory usage and nearly maxing out my page file. My usage generally has Firefox using between 300MB and 500MB or RAM depending on what I have open. The same usage on Chrome is tricky to track total memory usage, but I can get a rough idea of how much memory it was using when I quit Chrome and compare the available memory to how much memory was being used while Chrome was running.

What browser do you consider not to be a memory hog?

What platform do you run? I am on Linux x86_64 and Fx4b7 takes almost 10% of my RAM and ~20-30% CPU with two tabs open. With the same tabs open in Chromium, I get 2.7% RAM usage and 0-5% CPU usage. Maybe Fx has been really optimizing for Windows lately, but the performance on Linux x86_64 is not very satisfactory.

I am on OS X (10.6.5). I did not see FF4 CPU usage anything near what you are seeing. My CPU is ~2-3% with 6 tabs open (2 being Gmail). RAM usage is steady after a day of using the same browser process.

I used to be a die-hard firefox user and I also used to write extensions for it for fun and professionally as well. I always say it's the only browser that I've read the source code for.(well, some of it at least)

The thing that bugged me with FF3 was that it never seemed to reclaim the memory that the tabs were using after being closed. I noticed that when you have say >20 tabs open. I often had to save my tabs and kill FF so that it doesn't bog down my system.

Now I'm using chrome which still is not perfect especially when you have say gmail and gmaps open and several other tabs. But killing the tabs does seem to make the app snappier again.

The changing of the Panorama hotkey from ctrl+space to ctrl+e ist slightly annoying. I know that ctrl+k does the same trick and has been there for far more versions, but I've always used ctr+e for the search bar, mainly because that's the one that also works in other browsers. Of course I'm sure it will only take two seconds to find out how to change this back...

Browsers keep improving JS speed, but is there an upper limit to how fast it can be? How far off is it from something like C ?

JS speed has improved so much recently due to optimisations like JIT (Just In Time compiling). There will be an upper limit eventually but we haven't hit it yet.

It's damn fast but still a high-level interpreted language - time is taken for a program in the browser to compile it on the fly and run.

It wont hit the speed of well written C code, but most things don't need that speed anyway.

I was browsing the language shootout the other day and came across the page for the regex test. I'm not sure how it's doing it or if something is wrong with the test but V8 is outperforming everything else, including C. That's the only test it leads though.

Here are the results: http://shootout.alioth.debian.org/u32/performance.php?test=r...

Also worth noting that TraceMonkey isn't too far behind. Does anyone know why the JS engines, especially V8, are performing so well on this test?

P.S. Here's how V8/TraceMonkey stack up for all the tests: http://shootout.alioth.debian.org/u32/which-programming-lang...

V8 JIT-compiles regular expressions. I don't know what C/C++ do -- I'd assume they run the state machine directly, but maybe they do compile it further like V8 does.

http://blog.chromium.org/2009/02/irregexp-google-chromes-new...

Look at the source code of the C program. It uses glib and the TCL libraries and it even includes locking for the multicore benchmark.

Look at the source code of the other C regex-dna programs which don't use glib or Tcl libraries.

See the notes http://shootout.alioth.debian.org/u32/program.php?test=regex...

The Mozilla blog recently posted something about this, saying how all the major browsers are approaching the upper limit of JS speed. Can't find that post now though, but pretty sure it was a HN submission.

There might be an upper limit on the speed of the language itself, but what has a lot more room for improvement is speed of DOM operations and DOM rendering. Since you could involve all kinds of fancy graphics hardware acceleration in that, as IE9 is purported to do, there's a lot more to be done there.

Most JS developers agree now that the slowest operations are DOM operations so they are avoided as much as possible.

Re-introduces the PDF focus bug they made in FF 3.6. Beta 6 did not have this problem.

Nice work Mozilla.

Sometimes I wonder why we give so much importance for 1-2 second page load / javascript improvements. Does average user really care about 1-2 second difference?

Yes, certainly they do. Amazon even did that famous internal study showing that, other factors being equal, a page that loaded 200ms more slowly caused a statistically significant drop in revenue. (Even if end users aren't aware that they care, they care.)

Yes. Two seconds is usually regarded as the rection time threshold where an ineractive system starts to turn into a batch system. It is long enough for the user to start to forget what he expects from the system or to think about swiching to another activity while the current one loads. That is also the reason why database transaction benchmarks are always defined by number of transactions per second at a response time of slightly under two seconds, that is the threshold where they can still claim that the benchmark measures something about a system that is actually usable in practice.

So cutting response times by a second from say two seconds to one second changes the percieved delays from barely tolereble to barely noticable, which is unually beneficial for interactive apps.

My GOD yes. 400 milliseconds is big. 1000-2000 milliseconds is huge. Details from Google and Bing (via the ever-awesome Brady Forrest):

http://radar.oreilly.com/2009/06/bing-and-google-agree-slow-...

The "Web Console" is pretty awesome, but I don't see a way to close the detailed view things. Also I don't see POST contents in the detailed view things.

Thanks! Glad you like it. We've got some really nice improvements landing in beta 8/9.

You should be able to close the detail panel by using the OS-standard close control on the window. If you don't see that control, that's a bug! (The control certainly shows up on my Mac.)

Request/response bodies are not logged by default because they can be large. If you right-click on the output area, the context menu there has an option to log request/response bodies. I believe beta 7 is the one in which we put in a change so that only the first 1MB of the body is logged.

But does the user interface still respond like molasses in January? Last time I tried Firefox 4, I could have sworn my entire PC was locking up any time I clicked my mouse on any piece of the browser besides the web pages it was rendering.

No offense or anything, it was just an insurmountable obstacle, and without reasonable UI performance, render times and tab changes mean nothing.

I have begun testing, and I can report it is drastically improved.

Definitely feels snappier. I wonder if that's because it's missing some of my plug ins.

Nice, although I need Netflix & Firebug to work before I upgrade.

Firebug 1.7 alphas are compatible: http://getfirebug.com/releases/firebug/1.7X/

Congrats Mozilla on the release!

Odd title though for this post, for where is the "everyone"? The link shows comparisons to previous versions of Firefox, not other browsers.

One thing I hope they fix before final release though is the constant CPU usage when you have many tabs open. I thought the Panorama feature was supposed to address this. It's currently idling at around 25% CPU for me, and no, unfortunately I can't blame it on Flash. :-\

> Odd title though for this post, for where is the "everyone"? The link shows comparisons to previous versions of Firefox, not other browsers.

http://arewefastyet.com/

Sunspider’s not a very good benchmark; more interesting to me is that they’re nearly even with V8 in V8’s own benchmark.

Though I wonder why they can’t backport TM, or at least a subset, to 3.6; is Firefox so monolithic that improvements to the scripthost can’t be separated from the rest of the browser?

I ran the tests on my Ubuntu 10.10 laptop running Firefox-4.0 Beta8pre and Chrome 9.0.570.0 (64571) and Firefox actually beat Chrome 2943 to 3645 (or a 23% improvement) averaged over two runs.

Gotta love browser competition!

don't ever average software benchmarks, it doesn't make sense. always take the best run.

Huh... I haven't heard anyone in the Mozilla world even talk about backporting to 3.6 (even to say it's a bad idea). Everyone is very focused on 4.

I'm very happy with the result. However, on some javascript intensive webpages, such as https://console.aws.amazon.com/s3/home, it hangs the whole Firefox for a little while. Hopefully with a long timeline to the release, no such issues remained.

Err, your last sentence is a bit weird. First of all, 3.6 already has TraceMonkey; what's new in Fx 4 is JägerMonkey, and the integration between it and TM. Second, the Mozilla JS engine is largely developed separately (all of the actual development these days is in a separate repository, which merges semi-regularly with the main mozilla-central repo), but have you seen any other browser backporting anything other than security fixes?

>It's currently idling at around 25% CPU for me, and no, unfortunately I can't blame it on Flash. :-\

I sometimes notice this too, but always after running a bunch of Flash vids. Haven't methodically tested it, since it's not a priority and a quick restart fixes it. But how do you know it's not Flash?

Flash runs in its own process named "plugin-container", so if Flash crashes it won't take down the rest of the browser. I'm pretty sure that if flash were eating CPU, it would be attributed to the "plugin-container" process and not the "firefox" process.

All I want is a light weight FireFox .

Funnily enough, that's why we got Firefox in the first place---they wanted a lightweight (browser part of) Mozilla.

Of course, SeaMonkey is pretty fast nowadays too.

more robust they are trying to make it , more heavy it is becoming ...

Since they are ripping off the chrome UI part and parcel, I wonder why they still haven't switched to a unified search and location bar.

I still prefer the Firefox's Awesome bar over Chrome's unified bar. I like having the drop down list of my most visited sites.

There is a great Firefox plug-in called CyberSearch. I can't imagine browsing without it. For me, it works way better than the integrated location bar in Google Chrome.

Because not everyone thinks it is good UI?

Funnily enough, one of my few complaints about Chrome is that the location bar doesn't seem as good as the one in Firefox. I tend to use the location bar more often for accessing sites I've been to before, rather than searching for new sites.

Chrome doesn't appear to have a particularly good history search, so I have to type quite a lot before it figures out where I want to go. Conversely, Firefox seems to get it right more often, and with a lot less typing. For example, in Firefox, if I type in "y" in the location bar, it instantly suggests "news.ycombinator.com", which is by far my most frequently visited site with a "y" in it. Chrome suggests "yardoc.org", which I've only visited once or twice several weeks back.

Chrome seems to place a lot more emphasis on matching the address bar in order; so it assumes that if I wanted "news.ycombinator.com", I'd begin with "n". It also seems to rate sites based on how recently I visited them, rather than how often I visit them.

All this means that whilst the Firefox location bar seems to read my mind, the Chrome location bar hardly ever produces relevant results.

Reply on news.ycombinator.com