Firefox fixes save gigabytes of memory on image-heavy pages

https://blog.mozilla.org/nnethercote/2013/10/01/memshrink-progress-week-117-120/

by AndrewDucker • 13 years ago

440 193 13 years ago

blog.mozilla.org

icebraining2• 13 years ago

Great news! Many thanks to Timothy Nikkel and all the other Mozilla devs. It's particularly great on Fennec, where the memory constraint is noticeable.

For people - like me - who didn't know what B2G was, it's Boot2Gecko, the OS project: https://wiki.mozilla.org/B2G/FAQ

soapdog• 13 years ago

Check out http://firefox.com/os for a user view of Firefox OS and https://developer.mozilla.org/en-US/docs/Mozilla/Firefox_OS for a Developer view.

mccr82• 13 years ago

For people who don't know what Fennec is, it's Firefox for Android. :)

anentropic• 13 years ago

Or alternatively... Firefox fail was previously wasting gigabytes of memory on image-heavy pages

01Michael10• 13 years ago

Gee, the trolls are up early today...

morkbot• 13 years ago

On one hand true, on the other - it isn't that those fixes are some magic, extra smart hackery (at least on the idea level, dunno about the implementation). Don't load an image that user don't see, remove an image that user saw but don't see anymore - those seem as rather obvious solutions.

adamtulinius• 13 years ago

I wish people would use regular 2d plots for 2d data.

Other than that, great news!

eterm• 13 years ago

Technically it's 3D data, with the three axes being {Time, Memory Usage, Version}.

But I agree there may be a better way of visualising this data.

mattlutze2• 13 years ago

Technically it's three trials of 2D data, with the axes being Time and Memory Usage. The width of the lines given by the version axis is for visibility, not to encode data.

A 2D line graph with 3 colors for the different versions would be an appropriate visualization. This one suffers spatial distortion and is only useful in confirming that yes, they use less memory in step 2 now.

batbomb• 13 years ago

3D data, but 2D plots.

A histogram of a normal distribution (say, mean lifetime of a particle) is a 1D histogram. You would have counts of decays vs time. In your example, that would be 2D, but it's not.

A 2D histogram would be more like a heat map. The X axis would be time, y axis version, and the (x, y) value would be memory.

nnethercote5• 13 years ago

Everyone's crapping on me for the choice of graph. I originally tried a 2d plot but the lines obscured each other.

More importantly, it wasn't intended to be a rigorous data representation because the results vary significantly between runs. I was just trying to communicate the results in an approximate fashion; the improvements are dramatic enough that approximate is good enough, IMO.

And in case anyone is thinking about suggesting errors bars: please don't.

qznc• 13 years ago

I would suggest something like this: http://imgur.com/QugrU9y (inaccurate numbers!)

sfjailbird• 13 years ago

Sounds great - does anyone know how this compares to Chrome (both in terms of approach and results)?

sanxiyn2• 13 years ago

Chrome did a variant of this (at least) the last year. Not sure how comparable. See WebKit bug 94240 for example.

bru• 13 years ago

As spyder says in another thread of this post, on the given page (http://congressoamericano.blogspot.fr/p/fotos-do-congresso-a...) Chrome decodes and decodes... the memory usage goes up and up. I was using 1.5GO of memory before closing it (and still far from having loaded everything).

kamjam2• 13 years ago

Do you know if the tabs run in separate processes yet (like Chrome) or will a crash/memory leak/long running script cause the whole browser session to close?

nikbackm• 13 years ago

Good to see continuing improvements in the browser of the free and open net.

neurostimulant• 13 years ago

When loading the page, physical memory consumption would still jump to almost 3 GB, because the images are still decoded. But it would soon drop down to a few hundred MB, as the decoded data for non-visible images was discarded, and stay there (with some minor variations) while scrolling around the page.

This is a great news. The application that uses the most memory on my machine is the web browser.

I wonder if this fix would make scrolling the page less smooth.

nnethercote5• 13 years ago

I wonder if this fix would make scrolling the page less smooth.

It shouldn't. But why don't you try it?

mnarayan015• 13 years ago

Like many of the other Firefox memory changes, this one leaves me feeling a little ambiguous. On the one hand, obviously keeping memory usage down is good. On the other hand, waiting to decompress/decode images until they are visible to the user means that there are obviously going to be performance trade-offs (and also, obviously, design/engineering trade-offs).

Speaking just from my personal usage: I don't think there has ever been a time where delaying the image decompression on a page would ever have had any utility for me (at least from a memory standpoint). On the other hand, there have been plenty of times when eagerly performing image decompression would make my browsing a bit smoother.

All in all, it seems (from a not-particularly-in-touch outsider perspective) as if Mozilla went from being totally dismissive of memory complaints to being hyper-responsive to memory complaints with no (publicly stated) rationale behind the change. It would be interesting to see the reasoning behind this (if only from an intellectual standpoint), particularly given that in many cases I tend to feel that it would have been better to simply be a bit more diplomatic when rejecting memory-based bug reports.

dbaupp5• 13 years ago

I imagine it's good for image heavy sites like Pinterest or Tumblr.

mikebonnell• 13 years ago

Agreed, it's a huge win there.

daleharvey2• 13 years ago

With the capbility being built, if there is a performance tradeoff (I didnt look through the bugs, but imagine it was already thought about), then its now possible to do 'hey you have a load of spare memory' pre decode all the things type optimisations

mnarayan015• 13 years ago

Good point. I imagine the issues would be more with when to unload an image, though depending on the implementation that might not be particularly problematic either.

sanxiyn2• 13 years ago

https://wiki.mozilla.org/Performance/MemShrink has some reasons. One reason is explicitly "Fewer people will complain that Firefox is a memory hog and that Mozilla ignores memory usage". Even if complaints are technically nonsense, addressing complaints is still important.

mnarayan015• 13 years ago

Good link...thanks.

Even if complaints are technically nonsense, addressing complaints is still important.

Totally true, but I don't have to like it...and I can fight against it by complaining about the addressing of the complaints!

AshleysBrain• 13 years ago

If image decoding is asynchronous (don't know - guessing it is already?) and you decode them when they get near to the viewport and not just when they become visible, then it should always be decoded by the time you scroll to it and yet never jank the page. Scrolling a long way really fast probably means there's a small delay while it decodes images, but surely that's worth it to save gigabytes of memory?

mccr82• 13 years ago

Image decoding in Firefox was recently made multithreaded.

mnarayan015• 13 years ago

but surely that's worth it to save gigabytes of memory

I'm not sure _save_ is the right word here...using available memory has no particular cost to me over not using it. Of course, the _available_ modifier is relevant, so I take your point.

dkhenry• 13 years ago

I have been running aurora for a while so I already have this fix. I just tested it and there was no noticeable performance impact for the memory saved.

Chrome on the other hand used all my memory and caused me to start swapping. Any website with lots of images loaded will be impacted by this ( facebook , G+ , tumblr , ... ) so I think its a rather important change

akjj• 13 years ago

This was a follow-up to an earlier change which also delayed image decompressing. In the description of that change the author wrote: "Andrew McCreight wrote a comment that I can’t find now, but basically he said that on one of his machines, that page was extremely slow and painful to scroll through before the change, and after the change it was smooth and pleasant."

If you have any examples where the linked changes causes performance degradation, you should try reporting it to the developers.

smackfu• 13 years ago

I assume the old way choked very badly once you ran out of real RAM.

NathanKP3• 13 years ago

We've been using stuff like Infinity.js (http://airbnb.github.io/infinity/) for quite a while now, and it's basic functionality is to improve long scrolling panes by dynamically adding and removing elements (including images) in real time as you scroll to keep the number of elements low and the page responsive. Sites that use this technique usually feel just as responsive as if the site really did have a page with 10,000 items on it instead of a dynamic viewpoint that only shows 100 items as a time.

Personally I think it would be best if HTML developed a spec for an official UITableView style native element. That would provide a way to solve the memory issues and avoid the need for JavaScript hacks like Infinity.js

But in the meantime browser workarounds like this one from Firefox will help avoid the huge memory usage in sites that don't use smart techniques like Infinity.js

swift2• 13 years ago

I don't know that I'd characterize this as a workaround. Infinity.js is definitely a good approach, but you should expect good baseline performance from the browser even without more complicated techniques.

driverdan2• 13 years ago

The article states that it reduces initial page load time by about 4 seconds. Load time is far more important than a few ms here and there to load images on the fly when they are needed.

joosters• 13 years ago

Image decompression is pretty easy, CPU-wise. Well over a decade ago, computers were happily rendering JPEGs on-the-fly without storing a decompressed version (RISC OS apps did this). Your desktop/laptop/phone has plenty of ability to do this.

Lagged2Death• 13 years ago

...it seems ... as if Mozilla went from being totally dismissive of memory complaints to being hyper-responsive to memory complaints with no (publicly stated) rationale behind the change.

I don't know what their internal rationale is.

"Memory is cheap," we like to say, but it's really only true if you're not wasting it like crazy all the time. Ten years ago you could surf the web with 512MB of RAM; now we're getting to the point where 4GB isn't considered enough. If 4GB now costs the same as 512MB then, then in the sense of "dollars worth of memory required to perform function X acceptably," memory hasn't actually gotten any cheaper.

If you can get performance that used to require 4GB down into 1GB, you've upgraded the performance of a bunch of older and low-budget devices from "barely usable" to "fine."

Take the big-picture view: updates like this can actually result in a slower flow of "obsolete" computers and devices into the landfills. This is one of those bottlenecks where a little good engineering can have really disproportionate positive aggregate effects.

This sort of thinking has been one of the selling points of desktop Linux for a long time. I wish Microsoft thought more like this.

mindcrime• 13 years ago

Right. Also, no matter how cheap memory is, you run into limits of things like maximum RAM capacity. I'm there now... I have 8GB of RAM in this laptop, and that's all it can hold. I can't spend more money to put more RAM in here even if I want too, as it just won't take it. I'd love to go up to 16GB, but that won't happen until I buy a whole new machine.

Needless to say, I'm very happy about this firefox news. AS somebody who routinely runs a lot of apps (and more than a few memory hungry ones) I want every app to be a freaking memory frugal as it possibly can, and be a good citizen when RAM has to be shared between apps.

Playing the whole "Unused RAM is wasted, take up as much RAM as you can" mindset is fine if you only run one app at a time, but, really, who the f%!# actually does that???

dragonwriter4• 13 years ago

All in all, it seems (from a not-particularly-in-touch outsider perspective) as if Mozilla went from being totally dismissive of memory complaints to being hyper-responsive to memory complaints with no (publicly stated) rationale behind the change.

I suspect that being a company that wants to have a unified code base for their browser and that is now newly pushing a browser-centric OS for smartphones may explain this, at least in part.

nnethercote5• 13 years ago

I first started working on improving memory consumption in early 2011, and that evolved into the MemShrink project (https://wiki.mozilla.org/Performance/MemShrink) a couple of months after that. Firefox OS came later.

In fact, it's fortunate that MemShrink had been up and running for a while because it meant we had already fixed a lot of leaks and inefficiencies, and we had good tools for identifying memory problems in Firefox OS.

nnethercote5• 13 years ago

Have you tried the new version? You should try it before you dismiss it due to theoretical concerns. As I wrote in the blog post, the time before the page displayed dropped from 5 seconds to 1 seconds. And scrolling performance is unaffected AFAIK; if you have evidence otherwise Timothy and I would love to hear about it.

illumen• 13 years ago

Firstly... thanks! I love firefox, and appreciate any memory savings.

Did you do any tests for animation performance, and test that preloaders are not affected? Are these issues addressed?

Hardware accelerated decoding(like IE, and image viewers use) would help. Much modern hardware can do the whole decode-resize step in dedicated hardware.

For software only, an improvement is resizing in jpeg-space to the exact size required. It is much faster than decoding the entire jpeg (eg 5 megapixels 2MB jpeg, into 20MB of uncompressed image data) then only displaying it at 320x200 pixels(1MB at the 16bit needed for this YUV data). You can resize the smaller jpeg-space data, then not have to allocate that 20MB of data (perhaps never going back to the OS).

As well, some images compress very nicely with hardware texture compression (but many look awful). There are heuristics that can be used to decide the best type of texture to use based on the image data.

Why turn 6bit-16bit per pixel jpeg data into 32 bit per pixel data? It only takes more memory and is slower to draw. Surely jpeg usage is common enough to special case its handling?

Good luck!

spyder• 13 years ago

I just compared it with Chrome using the example page mentioned in article:

In Firefox Nightly at the initial page load it uses around 250 MB memory and with scrolling it stays around that.

In Chrome (in Canary too) initial it uses around 300 MB but scrolling keeps filling the memory indefinitely (I stopped scrolling above 1.2GB)

aktau2• 13 years ago

Yea, I also noticed that, save for the pages with tons of images, firefox has actually been better than chrome in that respect for the last few releases. Now with this fix, I'm seriously thinking of jumping back. Then I'll finally have my sweet DownThemAll again. Many kudos to the firefox devs, performance and memory-usage is very important, and it seems they realized that before it was too late. Even with my 16GB I'd really like to run an extra VM for development as opposed to letting my browser gobble it all up.

patrickaljord• 13 years ago

isn't 16GB enough to run an extra VM for dev? also, try docker.

Peaker10• 13 years ago

Last I tried Firefox, pretty recently, it still needed to address some serious issues, when compared to Chrome:

* Unresponsiveness due to one tab thrashing

* Bad use of screen-space (bookmarks always visible, cannot disappear titlebar like in Chrome [there's only a Windows extension for that], menu-bar takes up another bar of space by default)

* Launching firefox has stop-the-world behavior to check/update extensions -- interrupting your flow when you need your browser most

* Some extensions still require restarts

These are the ones I remember off the top of my mind, I'm sure there were more...

I really tried going back to Firefox, but the cumulative pain of all of these made me go back to Chrome.

zobzu• 13 years ago

im using windows and the regular firefox (so it doesnt have this fix) and i notice that chromium always scrolls a lot slower than firefox anyway. which annoys me hehe.

uladzislau• 13 years ago

Chrome became very resource hungry and poor at memory management recently. Sometimes I see half blank pages when scrolling which I've never seen before on the desktop.

meowface• 13 years ago

Yep. In fact, I've noticed performance appearing to get worse and worse in both my Canary browser and regular Chrome. Canary always seems to perform quite a bit worse than regular Chrome. Not sure if that's just due to bugs because of the unstable build, or further changes they're making that is continuously hurting performance.

Zr40• 13 years ago

The half-blank pages you're seeing is a feature that's recently been enabled. This is a case of latency improvement that visually appears to be worse.

When rendering can't keep up with your rate of scrolling, Chrome will display the part that has been rendered so far and display a checkerboard pattern for the not yet rendered part. Previously, Chrome would just flip the display buffers less often if rendering can't keep up.

vectorjohn• 13 years ago

This has been a myth floating around since the early days of Chrome. I am and have for a long time been a big fan of Chrome, but it has never been light weight. It's just fast so people think of it that way. But you've always been able to check on the memory usage of tabs and see it just eating up your memory.

ohwp• 13 years ago

Opera 12.16 kept it's memory around 330 MB (initial 270 MB).

mgraczyk• 13 years ago

Interestingly enough, I loaded the page in IE 11 (11.0.9600) and scrolled all the way to the bottom. It initially used around 160 MB, then reached 240 MB by the bottom of the page.

Chrome used around 2.5 GB on my machine.

reidmain• 13 years ago

Is it time to switch back to Firefox from Chrome? I remember when I originally switched around Firefox 3.5.x Chrome was much leaner and faster. Has the pendulum finally swung back the other way?

dbaupp5• 13 years ago

It depends how you use it: e.g. I normally have many many tabs open, and Firefox handles that much better on my system than Chromium.

culshaw• 13 years ago

really? I have about 20 tabs open on Chrome now with no lag, with one tab open in Firefox my whole system grinds to a halt.

shmerl5• 13 years ago

Yeah. Many tabs cause Chrome to spawn many processes. It becomes a memory hog eventually.

Derbasti2• 13 years ago

For me, it is time. I switched back about a year ago, when Firefox started to not be painful any more, but still was a step back from Chrome. Some tabs would still cause stuttering in the whole app and sometimes scrolling was less than smooth.

These days however, I have not noticed any of these things for months in Firefox. Really I can't recommend it enough!

cpeterso2• 13 years ago

For me, it is time. I switched back about a year ago, when Firefox started to not be painful any more, but still was a step back from Chrome.

What was the tipping point that prompted you to switch back to Firefox? Was Firefox able to import your Chrome history and bookmarks without losing anything?

snarfy• 13 years ago

Firefox will never ask you to sign-in for the best browsing experience like chrome does. That's when I switched back. You can tell google is an ad company and users are secondary by their default choice of 3rd party cookies. I'd rather not have an ad company developing my browser. The incentives are all wrong.

swang• 13 years ago

While Firefox isn't an ad company, don't they get most of their revenues from Google's placement as the default search engine in Firefox?

driverdan2• 13 years ago

I've been using FF lately for cross browser testing. Until tabs run in their own process / thread I won't be switching back. Also, while the dev tools have come a long way and are pretty good now they're still not quite as good as Chrome.

shmerl5• 13 years ago

I never saw a reason to switch to Chrome to begin with. Performance was acceptable for me, while other benefits always made Firefox more useful (privacy with encrypted Sync, better add-ons and etc.). And some time ago, Firefox clearly became a leader in performance as well.

All this "jumping to Chrome" was simply a hype thing IMHO, and I generally don't care about hype.

syncerr• 13 years ago

From a developer standpoint, the dev tools in Chrome far exceed those in Firefox. The nightly releases are just starting to compare.

sliverstorm4• 13 years ago

There was a period around FF14 where FF was performing terribly for me, and Chrome was acceptably fast as ever. That was what made me switch for a while.

toki5• 13 years ago

"Reasons that don't necessarily apply to me," while fair, do not fall under "hype."

hrjet• 13 years ago

I use both, simultaneously.

I have 4 different Chromium profiles, and two different firefox profiles, with different levels of privacy / security. Each profile is also linked to a particular context, for example, a tab each for project A, project B, personal, etc.

swinglock• 13 years ago

Absolutely. The only thing I miss from Chrome is the clever indicator in the scrollbar when searching on the page. The only thing.

nnethercote5• 13 years ago

Try it! It's not hard. And if you do, please reset your profile first: https://support.mozilla.org/en-US/kb/reset-firefox-easily-fi.... Lots of old Firefox profiles have problems that can hurt performance, and resetting can make a big difference.

PedroBatista• 13 years ago

Excellent news! This problem was so bad that after only a couple pages of porn gifs on Tumblr i had to open Chrome to keep the smut going.

warcode• 13 years ago

I returned to Firefox (Aurora) a couple of months ago, but i am currently starting to inch towards chrome again.

While reduced memory usage is good, it doesn't help when has major problems handling multiple open tabs with flash player, has problems with javascript-heavy pages, and the firefox profiles STILL degrade in performance over time.

I really don't want to remake my FF profiles every 2 weeks just to avoid performance issues.

cpeterso2• 13 years ago

What Firefox add-ons are you using? You shouldn't need to rebuild your profile every two weeks. How much does rebuilding your profile improve performance?

rplnt• 13 years ago

Tried the page with Opera (12.xx) and it peaked at around +850MB and within seconds backed down to the initial value (+0MB). No change during scrolling. How is that possible? Does the browser hold bigger pool and just got rid of other stuff?

Sarkie• 13 years ago

Chrome is starting to be a nightmare to use, might have to go back to Firefox.

I can't get meaningful help if it freezes or doesn't load and it can't deal with 50% of the router pages, IE only :(

ksec• 13 years ago

For those Complaining about Firefox Responsiveness. There are e10s Reboot ( Process Per Component or tab ) that they are working on. Generational GC, New Graphics Backend etc.

acdha• 13 years ago

Interestingly, this optimization is already unnecessary if you are using the common technique of setting the CSS background-image on an element rather than using IMG src - on Chrome, Firefox and even Internet Explorer memory usage will be stable on a grid of images that way and, as a bonus, you trivially get cross-browser scaling and centering for vertically scaling thumbnails to be the same size.

Avitas• 13 years ago

Nice. This means that SeaMonkey will be getting these updates a day or two after FF.

chestnut-tree• 13 years ago

Completely unscientific test: if I load the following website in Firefox 24 (on Windows 8), the page scrolls very sluggishly (as it did in previous Firefox versions). The same page scrolls fine in Chrome. I don't know what is causing the difference. Is it the images? Or something else? For info, Firefox is my main web browser and I have never switched to Chrome (nor do I intend to).

Warning - the website is ridiculously image heavy (it's a gallery of Android app screenshots)

http://androidniceties.tumblr.com/

trustfundbaby• 13 years ago

If Firefox would find a way to let Chrome extensions work in Firefox, that would definitely make it a little easier to switch back and forth, definitely going to try this out though.

mark-r• 13 years ago

I switched to Chrome recently because Firefox was growing overnight to the point where it started thrashing and becoming unusable. I just checked and it says I already have the latest version. This recent change must not be the whole story.

what_ever• 13 years ago

The latest changes done as part of this are only present on Aurora (Firefox alpha) and the feature is turned off by default.

Also, I have my Firefox (Aurora again) running for over a week now on my dev machine with > 50 tabs open an I have had no issues with memory. I suppose there is some problem with one of your add-ons. Try creating a clean profile and see if it's any different.

Firefox fixes save gigabytes of memory on image-heavy pages

Related Stories