Release The Kraken
blog.mozilla.comHere are my results (Win7 64b, Core2 Duo 2.4 GHz):
Minefield 4 b6 9737.7ms
Opera 10.62 14490.8ms
Chrome 7.0.517.5 dev 18375.4ms
Firefox 4 Beta 5 21721.1ms
Safari 5.0 22168.6ms
Firefox 3.6.9 30053.3ms
Explorer 9 PP4 64817.2ms
What's probably more interesting is the order of other browsers (for which this benchmark should be less tuned) - again Opera performed very well (and better than Chrome).
My results (OSX 10.6, Core2 Duo 2.4GHz)
Safari 5 13939.9ms +/- 1.6% Chrome 6.0.472.55 19155.2ms +/- 1.3% Firefox 3.6.4 23567.1ms +/- 2.8%Yeah, on my MacBook Pro (OS X 10.6, Core i5 2.4GHz) I got
Interesting (but maybe not surprising) to see that Safari is consistently being reported as faster than Chrome on OS X, but not on Windows.Safari 5 11018.2ms +/- 0.3% Chrome 6.0.472.55 15342.7ms +/- 1.0% Firefox 3.6.8 18917.6ms +/- 0.1%My iMac with Safari 5 was a little more 14ms...
I see what's going on here. Mozilla's trace compilation approach has a lot of startup overhead but produces extremely fast optimized code for inner loops. Therefore, their benchmark includes compute-intensive tests with long-running inner loops to amortize out the costs of JIT compilation and focus on the raw speed of the generated code, where they beat Chrome.
(Note that I'm not saying this is unfair or biased at all. This is a good set of benchmarks to have.)
I think JS performance on Firefox has now reached the point where its good enough. All I ask for now is parts of the UI to become multithreaded so as to prevent the random issues one gets when having in the neighborhood of 150 tabs.
I don't think there's a software solution for the issues someone with 150 tabs open must have.
Yes there is. It's called Google Chrome. I presently have 164 tabs open in it across 6 windows. It works just fine.
I am regularly amused by the stream of inane blog posts from "UX Professionals" declaring bankruptcy because oh god they have 22 tabs open — and then they go on to propose some ridiculous unbuilt UI that's even less capable, but makes them feel designery.
Chrome does start to bog down eventually, and I can respond to that by killing off worker processes, which leaves the swath of tabs it was responsible for dead, but still in place with their URL and I can just hit refresh. With Firefox the whole browser will repeatedly lock up for seconds at a time at a much lower usage threshold and eventually crash completely, taking 5 minutes out of my day. Before I switched to Chrome, I would have Firefox would crash at least 10 times per week. That and their regular releases baking AwesomeBullshit into Firefox instead of releasing extensions was more than enough to compel me to ditch it.
I still don't understand why you could possibly need 164 different web pages open at the same time. What kinds of things are you doing that simple bookmarks can't solve?
They form stacks of things that haven't been finished with yet. If I bookmark something in that state, it just falls off the end of my brain. In a tab it remains in my periphery.
I shall attempt to document what I am presently filling all those tabs with:
Depending on what I'm doing at any one time, any one of the things counted in dozens could dominate. If you'd caught me during a RSS binge, there'd also ba a whole bunch of extra small windows each containing a single flash video embed, generated by the "Popout" functionality in Google Reader.* Personal Gmail and Google Reader, along with a dozen or so links opened from them * Several work-related OWA mailboxes and Google Apps mail and docs accounts * 10 Hacker News discussions, including the one I'm making this comment in. Normally there would be two for this discussion as I open the 'reply' link in a new tab to not lose my place, but this tab was from a Notifo growl * A couple dozen tabs are for music, TV, games, and films to investigate/pirate * A dozen profiles of people to get in contact with on social networking sites * Another dozen active threads on several phpBB forums I am a member of * Several dozen tabs of product pages from manufacturers, alibaba, and ecommerce sites for several physical projects I'm working on. * A dozen tabs of research for software projects I'm working onOf the tabs presently open, about half are new in the last 24 hours. Chrome's chrome://history/ page only lets me page back through 450 items chronologically, which is about a third of my average daily usage.
I've tried a number of tools to collect links in — bookmarks, social bookmarks, Google Docs, Google Wave, etc., but I always end up falling back to tabs because I'm already using them anyway! I'm currently giving the Chrome/GDocs bookmark syncing a shot for things that are 'out of mind'.
Do you only have one computer? Is it a notebook or desktop? It's hard for me to relate to these kind of use cases, because I use many computers, several of which are notebooks. I could never use a workflow like you because I'd always be losing all that state.
Right now I use a 27" iMac at home almost exclusively, but there's a handful of other computers that I use in various locations, and several have persistent sessions (which makes gchat think I am available almost 24/7). It was more of a problem when I commuted every day, but I mostly just had different sets of things that I was looking at on each compy.
The Google Bookmark sync actually works pretty well between all of them for things that aren't transient. I just realized that there's probably a decent 'bookmarks sidebar' extension out there, and combined with "Open All Bookmarks in New Window" that could alleviate a lot of my project-related tabcruft.
I have indeed "lost all that state" many times over the years, but it hasn't happened since I switched to Chrome (It did corrupt its session once, but it keeps a backup). It happened every couple months with Firefox on Linux, though I couldn't dig as big of a hole since it crashed more than once per day on average.
Years ago when I used Safari as my main browser it didn't have session saving yet, I used a SIMBL extension that continuously saved all state (including the full DOM with no redownloading), but ironically caused it to crash much more frequently. In those days I would rotate between Safari, Camino, and Shiira using two at a time, so that when the primary started to fall over I could start a fresh session and let the tabs dwindle in the old ones until they could be quit.
Re: my usecase, I have a 26" 1920x1200 monitor on my workstatop, with a 15" Powerbook (running Debian, 1ghz G4) next to it that is basically dedicated to "always online" stuff like IRC, IM, and Identica, but never runs anything else and never moves.
Firefox 3.5.x is slower than hell on it, and I last time I tried, Chromium wouldn't build on PowerPC.
Having two heads on seperate machines comes in rather handy for managing workflow of concurrent complex tasks; as in, I can yell at someone on one while doing work on the other.
I use synergy to "transport" input to the laptop.
Thats similar to how I do it. I have a virtual desktop for every task (with desktop 1 being my things without tasks/screwing around group), with at least one Firefox window open in every desktop....
Then, I have a Firefox window open for every subtask[1], which for example, could be a window full of documentation tabs, a window full of more documentation tabs for a different thing, and then a window full of even more documentation tabs for a third thing, all next to Eclipse, with a few vims open too for stuff I dont want open in Eclipse.
[1] I've actually started consolidating sub-task windows into tab groups inside Panorama/Eyecandy so its just one window per task. Control-space to bring up the local tab groups is very nifty.
I can believe you'd want 100+ tabs open, but I can't believe that flash wouldn't have forced a crash with even a dozen long running instances, unless you have the build with the new integrated flash plugin.
Chrome does a pretty terrific job of surviving the Flash plugin — it doesn't let it get out of control nearly as much, when it does it only affects a few tabs, and if it dies or I kill it everything's peachy with black boxes where flash once was. It shits the bed a lot more on Linux by a large factor, but Chrome's countermeasures work just the same.
Flash video playback does stutter regularly on my ridiculously overpowered machine. Occasionally a group of tabs will pause from an obviously flash-related issue (the flash open file dialog has been causing me grief), but it recovers in a couple seconds. Once a month or so I have to kill the flash plugin process because it goes on a runaway malloc train, but it doesn't affect my tabsplosion at all — I just have to refresh a couple of them when having flash loaded again would serve some useful purpose.
FF trunk, and now 3.6 (after they backported it), runs all plugins in isolated processes. I can manually killall plugin-container if flash goes nuts (like using over a gig of memory, which I've seen it do).
A lot of the icky problems with having a lot of tabs went away soley due to that (at least, for those without habitual flashblock usage).
It's very common when doing bibliographic work for research.
TV Tropes. (If you don't know what it is, it's a massively crossed linked wiki. Don't go looking for it, you'll lose many hours of your life.)
It could be worse, you could have told them about [REDACTED].
This isn't a problem with 150 tabs. An example of this problem is the firefox GWT developer mode plugin. When your app is compiling in this mode, which can take upwards of 30 seconds, the entire browser, including all tabs and windows, locks up. If for some reason the plugin screws up, you have to kill and restart the entire browser.
Firefox4 is faster than chrome here on ubuntu lucid x86_64 on an i5 with 8g of RAM.
Results for firefox 4 beta7:
http://krakenbenchmark.mozilla.com/kraken-1.0/results.html?%...
Results for chrome 7.0.517.0 dev:
http://krakenbenchmark.mozilla.com/kraken-1.0/results.html?%...
The beat detection section was most interesting to me, as I spent a lot of time trying (and mostly failing) to do reliable beat detection when I was in high school for an XMMS plugin. It amazes me that this is now possible in a web browser.
It'd be nice if one of the JS benchmarks actually POST'd the results back to the server, with current browser, and had a UI that allowed comparing the results for different browsers on different tests.
Both Kraken and Sunspider provide results in a form that you can copy and paste, and a UI for comparing results from different browsers. (But they don't let you view results from previous users, which I guess might be what you're asking for.)
Right, I'm asking for a way to see results in aggregate. Admittingly there's a problem of differing hardware (which I somehow overlooked).
Also, given the number of 'fans' on the Internet, I suspect that such a tool would have serious issues with users tweaking the benchmark and uploading fake data.
On mine, I notice Opera performs the worst by more than double.
(Win7, AMD Athlon XP2400 2.0GHz)
Chrome 6.0.472.55 beta: 46946.5ms
Opera 10.62: 234444.6ms
IE9: 71864.8msThose results would be roughly similar to the other poster's if you removed one of the four 4s in the Opera results. Maybe just a typo?
Not a typo (hopefully the links don't break): http://krakenbenchmark.mozilla.com/kraken-1.0/results.html?%...
http://krakenbenchmark.mozilla.com/kraken-1.0/results.html?%...
Of course it performs much worse on this benchmark than other browsers on my old hardware, but I haven't noticed any problems while browsing.
Is there any documentation? What are the components of this benchmark?
If you run the benchmark, the results include a list of the tests with links to short explanations.
EDIT: Click to the results page and explanation links: http://bit.ly/cXTBUb (using a redirect because HN broke when I used the real URI)
Here's the list:
ai: 922.9ms +/- 3.2% astar: 922.9ms +/- 3.2% audio: 5564.5ms +/- 3.4% beat-detection: 1996.1ms +/- 5.2% dft: 821.8ms +/- 6.8% fft: 1965.6ms +/- 6.0% oscillator: 781.0ms +/- 0.9% imaging: 2875.9ms +/- 1.4% gaussian-blur: 1271.2ms +/- 1.1% darkroom: 510.1ms +/- 0.7% desaturate: 1094.6ms +/- 4.0% json: 365.9ms +/- 1.6% parse-financial: 235.5ms +/- 1.6% stringify-tinderbox: 130.4ms +/- 2.1% stanford: 2392.1ms +/- 3.2% crypto-aes: 468.7ms +/- 2.0% crypto-ccm: 656.2ms +/- 3.5% crypto-pbkdf2: 1063.0ms +/- 4.5% crypto-sha256-iterative: 204.2ms +/- 2.9%Looks like a heavy focus on compute-intensive tasks. A* is definitely worth optimizing since it will help games.
Click on any of the components and it will explain the benchmark and give the source (which is heavily documented).