e10s / Electrolysis

[weamish]

I very rarely use multiple windows myself, but I was thinking that if separate windows were in a separate process that I personally would use them more, but not for my main browsing.

I use separate windows quite often, to facilitate flipping back and forth for chats etc.

[Tharn]

It would seem like a half-measure to me. I don't really see the advantage here. Using separate windows is already possible without performance issues. If you do get issues, that's because of heavy websites. They won't magically get any lighter; it'll just end up being one unresponsive window instead of both.

IMO, If you were going to split the browser into separate (few) pieces, there's only one set-up that'd make sense: Make one piece the user interface, the other piece the website rendering. Just two instances. And then have a long, hard look at how Opera 12 did it as far as the rendering goes. A website that wasn't actively displayed didn't impact performance with the Presto engine.

Of course this is still a massive undertaking and probably beyond a tiny dev team in scope. For something more realistic, I would simply hope that Palemoon continues to develop in a way where those bottlenecks causing websites to freeze the entire browser will be found and improved upon. I'm quite fine with a single-process browser.

[Gary5]

it'll just end up being one unresponsive window instead of both.

I remember some versions of Firefox when I would've killed for only one unresponsive window. But I haven't had problems like that in a long time, maybe because of using uMatrix.

[Agent Orange]

[…]Chromium UI is always snappy, 'classic' Firefox UI chokes on a single heavy website. […] Can someone else chime in here?

[…]All I know is that since those changes in both browsers, I can load as many tabs as I want without issue, a problematic tab doesn't slow or crash any of the other tabs, and resource usage has gone way down because tabs can be loaded and unloaded from memory so inactive tabs are unloaded until needed. I've personally found it to be a game changer for me.

Thirding this. The performance difference between Chromium and Firefox/Pale Moon is immense. On Pale Moon, the whole browser — scrolling, UI, etc, for all tabs across all windows just completely locks up for several seconds if I open e.g., an imageboard catalog or equally intensive page (even in a background tab!)

Performance is one thing I seriously miss about Chromium. The browser always behaved snappily, whether I had 3 tabs or 15 open. The UI always worked and was responsive, tabs could usually even be scrolled on if their JavaScript was messed up. The browser always just worked. And this, even on a significantly underpowered machine (single-core processor with DDR2 RAM).

If Firefox has as much success in its "Electrolysis" project as as Chrome has had with its multithreading, I may have to switch back to Firefox. I'll be sad to go; I appreciate all the improvements Pale Moon has, and Firefox's NTP Ads are simply unnerving. But this non-Chrome-like behavior leads to a browsing experience which can only be described as crippled compared to the reliable smoothness and responsiveness provided by Chrome-like browsers.

[megaman]

Performance is one thing I seriously miss about Chromium. The browser always behaved snappily, whether I had 3 tabs or 15 open. The UI always worked and was responsive, tabs could usually even be scrolled on if their JavaScript was messed up. The browser always just worked. And this, even on a significantly underpowered machine (single-core processor with DDR2 RAM).

Right now, I have 47 tabs open and hardly hurting my RAM, but on Chromium(SRWare Iron, Comodo Dragon, or even TheMapleStudio's ChromePlus(CoolNovo)), over 45 and I couldn't handle the browser's hang.
I'm sorry you feel that way? Just because we aren't moving towards a direction you want, our browser is undermined?
I have nothing but smooth experience on the Stable PM browsers, but Chrome doesn't make what browsers should be.
(Presto)Opera was Single-Threaded and had extremely less amount of resources used, while being "smooth and responsive."
In any case, this "smoothness" varies, I don't find Chromium any different than PM, you just want us to go by your beliefs and make it so, that's jingosim/chauvinism. I'll admit that I can't compare with the Single Core PC, since I don't have one lying around.

"I'll be sad to go," I don't believe that at all, it's just sounds like a pity. (I'm actually getting ticked-off with this "trump card" that you are playing, very insulting)

[Axiomatic]

Thirding this. The performance difference between Chromium and Firefox/Pale Moon is immense. On Pale Moon, the whole browser — scrolling, UI, etc, for all tabs across all windows just completely locks up for several seconds if I open e.g., an imageboard catalog or equally intensive page (even in a background tab!)

Performance is one thing I seriously miss about Chromium. The browser always behaved snappily, whether I had 3 tabs or 15 open. The UI always worked and was responsive, tabs could usually even be scrolled on if their JavaScript was messed up. The browser always just worked. And this, even on a significantly underpowered machine (single-core processor with DDR2 RAM).

If Firefox has as much success in its "Electrolysis" project as as Chrome has had with its multithreading, I may have to switch back to Firefox. I'll be sad to go; I appreciate all the improvements Pale Moon has, and Firefox's NTP Ads are simply unnerving. But this non-Chrome-like behavior leads to a browsing experience which can only be described as crippled compared to the reliable smoothness and responsiveness provided by Chrome-like browsers. :thumbdown:

Performance in the sense you are speaking is highly subjective. Realize multi-process will likely kill heavy tab usage for those who do no have more than 8 GiBs of RAM. When I ran Chromium, 2channel or anything involving loading many images would cause my xsession (or something related to it) to kill the offending task that was memory hogging since it was bringing my system to freeze up.

If Firefox has as much success in its "Electrolysis" project as as Chrome has had with its multithreading, I may have to switch back to Firefox.

Multi-threading is not the same as multi-processing.

Pale Moon does not have to switch to a multi-process model, rather it should not. Look ahead to Mozilla's Servo project, native parallelization is the main goal of it, why? Namely because parallelization is a fad, secondly because parallelization allows advantages on systems that are rich in hardware, but does not punish those who cannot afford the latest hardware.

Not to say Pale Moon should be re-written, largely because that task is simply too large; however, Pale Moon can take parallelization further than the auto-parallelization it currently uses. The auto-parallelization is not even 1/10th as powerful compared to implementing OpenMP pragmas through out.

Try setting OpenMP pragmas around Pixman, SVG, JPEG, QCMS. (Setting OpenMP for the for loops is the easiest way to get started.) I caution messing with OpenMP and Pixman, currently when I do I get segmentation faults.

TL;DR good things come to those who wait.

[Moonchild]

If Microsoft were to actually maintain and support OpenMP more fully, it would be interesting; but they don't.
I do have the plan to put in similar pragmas for auto-parallelization, when time permits. Even so, just setting OpenMP or other pragmas doesn't "magically" make code parallel or able to be made parallel by a library. This will have to be rewritten, and a good number of things simply can't be made parallel in their current code/method design. See also your segfaults as a result.

Also, heavy image use causing spikes in CPU and memory will be addressed in Pale Moon 26; it's a known shortcoming of the current release versions. That, though, has very little to do with multi-process setups, apart from being able to be lazy in your browser design and just offloading a task that might bog down your main thread to a separate process and letting that max out your core as much as it wants. Of course once you hit a number of tabs that all hit some resource snag in a multi-process browser that cares less about doing this right, then your entire system will choke, not just the browser

[Agent Orange]

Thirding this. The performance difference between Chromium and Firefox/Pale Moon is immense. On Pale Moon, the whole browser — scrolling, UI, etc, for all tabs across all windows just completely locks up for several seconds if I open e.g., an imageboard catalog or equally intensive page (even in a background tab!)

Performance is one thing I seriously miss about Chromium. The browser always behaved snappily, whether I had 3 tabs or 15 open. The UI always worked and was responsive, tabs could usually even be scrolled on if their JavaScript was messed up. The browser always just worked. And this, even on a significantly underpowered machine (single-core processor with DDR2 RAM).

If Firefox has as much success in its "Electrolysis" project as as Chrome has had with its multithreading, I may have to switch back to Firefox. I'll be sad to go; I appreciate all the improvements Pale Moon has, and Firefox's NTP Ads are simply unnerving. But this non-Chrome-like behavior leads to a browsing experience which can only be described as crippled compared to the reliable smoothness and responsiveness provided by Chrome-like browsers.

Performance in the sense you are speaking is highly subjective. Realize multi-process will likely kill heavy tab usage for those who do no have more than 8 GiBs of RAM. When I ran Chromium, 2channel or anything involving loading many images would cause my xsession (or something related to it) to kill the offending task that was memory hogging since it was bringing my system to freeze up.

Right now, I have an AMD E2-1800 processor computer (low end) with 4GB of RAM. While drafting this message, and with no other tabs open, I opened https://8ch.net/tech/catalog.html in a new tab, causing the browser to lock up for close to 10 seconds. ("Lock up": UI stops responding, all actions are queued/delayed until such time as the browser un-locks. About 4 out of 5 times, Windows realizes that the browser is locked up and appends Not Responding to the window title.)
As a direct comparison test, I pulled up Google Chrome while drafting this. I pasted https://8ch.net/tech/catalog.html into the URL bar, then proceeded to open a few tabs and Google things while the catalog was loading. At no point did the browser lock up or even scrolling give out.

Total Chrome RAM usage with the /tech/ catalog, 4 Google pages, and a wikipedia page (adding all the processes up): 160MB.
Pale Moon hovers around 278MB (though I will admit I have Adblock Latitude, Encrypted Web, and Greasemonkey installed, so it's not a fair comparison.) (The locking up behavior has been tested without add-ons, as well.)

So there appears to be very little room for subjectivity with regards to performance.

If Firefox has as much success in its "Electrolysis" project as as Chrome has had with its multithreading, I may have to switch back to Firefox.

Multi-threading is not the same as multi-processing.

I may have messed up on the terms here; I'm not entirely clear on the implications of multiprocess-vs-multithread. What I am saying is that whatever Chrome is doing is working, and working so well that Firefox is spending significant development time following suit.

Pale Moon does not have to switch to a multi-process model, rather it should not. Look ahead to Mozilla's Servo project, native parallelization is the main goal of it, why? Namely because parallelization is a fad, secondly because parallelization allows advantages on systems that are rich in hardware, but does not punish those who cannot afford the latest hardware.

As a low-end hardware user (slow processor and limited RAM), I stand to benefit the most from these performance boosts.

It's purely anecdotal (I haven't run any hard scientific-type tests on this yet), but in my experience, Chromium on a Pentium 4 runs about as fast as Pale Moon on a Core 2 Duo.

Whatever Chrome is doing performs excellently, even on low-end hardware. What Pale Moon and older Firefox versions are doing, by comparison, is NOT running on low-end hardware very well. The only behavior "punishing those who cannot afford the latest hardware" is that which Moonchild is currently planning to proceed in.

I understand if supporting hardware regarded as obsolete is not a project goal, however.

[__NM64__]

AMD E2-1800 processor computer (low end)

I also have a dual-core Bobcat-powered processor, so here's a better idea of how it's CPU cores actually perform...

At 1.6GHz, it is about equivalent to the following...

1.2GHz Athlon 64 x2
1.2GHz Core Duo (essentially a dual-core Pentium M)
1GHz Core 2 Duo (Conroe)

From there you can use the equivalent performance for the Athlon 64 and Core 2 Duo to compare to most other CPU architectures:

[megaman]

I also have a dual-core Bobcat-powered processor, so here's a better idea of how it's CPU cores actually perform...

Please, hide the image as it takes too much real-estate for the comment area.

Current laptop with the A6-6310(Quad-core 1.8GHz w/Turbo Core 2.4GHz), which I call a snail.
Comparing Maxthon, Yandex Browser(Chromium-based), and Palemoon, Palemoon starts-up last but loads pages faster, all with add-ons/extensions except Yandex.
Yahoo! = Palemoon and Yandex load equally fast. Youtube = Palemoon loads faster, then Maxthon. Capcom = Palemoon, then Maxthon. Mamma.com Palemoon then Maxthon.
I forgot that I was using the BETA version of Palemoon, might be different results from the current Release version.

Off-topic:
Unfoturnately, my old laptop has the E2-1800, which I gave to my sister, had the charger got bit by a mouse, so I can't use it.(Been sitting on her shelf for over a year, anyways)

[Moonchild]

Right now, I have an AMD E2-1800 processor computer (low end) with 4GB of RAM. While drafting this message, and with no other tabs open, I opened https://8ch.net/tech/catalog.html in a new tab, causing the browser to lock up for close to 10 seconds. ("Lock up": UI stops responding, all actions are queued/delayed until such time as the browser un-locks. About 4 out of 5 times, Windows realizes that the browser is locked up and appends Not Responding to the window title.)

I compared both side by side with a blank profile. Google Chrome indeed totals to 160MB (6 processes). Pale Moon after performing the exact same steps uses 145MB (single process).
Speed/performance while loading the 8ch catalog was exactly the same, no noticeable difference. Even with my system being faster than what you used, a hang like that would have been very noticeable. That is with Pale Moon 25.8.0 release.
So, unless you're counting a longer running session that will actually use more memory (by design) to cache rendered pages etc., the memory usage of Pale Moon certainly rivals that of Chrome with a low number of tabs. With a high number of tabs, Chrome loses quickly, and noticeably impacts overall system responsiveness by piling on more and more processes. For example, opening 2 new tabs of slightly more complex sites in both browsers, one to www.thesecretworld.com and one to www.bbc.com grew memory in Pale Moon by 20MB. Chrome spawned 2 new processes of 47+36=83MB total.

[megaman]

The 8ch site, Palemoon loads a bit faster, then maxthon, then Yandex.
There is a minute hang that dissipates rather quickly on Palemoon.

[__NM64__]

Please, hide the image as it takes too much real-estate for the comment area.

A measly 450x448 image takes too much real-estate? O_o

My desktop resolution is only 1152x864 and yet that image is on the smaller to medium side of things for me...

Current laptop with the A6-6310(Quad-core 1.8GHz w/Turbo Core 2.4GHz), which I call a snail.

That's because that uses Puma cores, which has the same IPC as Jaguar which itself is the successor to Bobcat.

For reference, Puma and Jaguar's IPC is only around the level of the Windsor Athlon 64 x2.

[megaman]

For reference, Puma and Jaguar's IPC is only around the level of the Windsor Athlon 64 x2.

Yeah, I only get refreshes. I'm waiting for Carrizo's refresh. Yup, my APU is Puma+. I find it rather insulting that it performs around Windsor's level.

At least it performs better than Bobcat.

So that I don't stray-off from the topic:
The Secret World: Palemoon, Yandex, Maxthon
BBC: Palemoon, then Yandex and Maxthon tied.

[Axiomatic]

Right now, I have an AMD E2-1800 processor computer (low end) with 4GB of RAM. While drafting this message, and with no other tabs open, I opened https://8ch.net/tech/catalog.html in a new tab, causing the browser to lock up for close to 10 seconds. ("Lock up": UI stops responding, all actions are queued/delayed until such time as the browser un-locks. About 4 out of 5 times, Windows realizes that the browser is locked up and appends Not Responding to the window title.)
As a direct comparison test, I pulled up Google Chrome while drafting this. I pasted https://8ch.net/tech/catalog.html into the URL bar, then proceeded to open a few tabs and Google things while the catalog was loading. At no point did the browser lock up or even scrolling give out.

Total Chrome RAM usage with the /tech/ catalog, 4 Google pages, and a wikipedia page (adding all the processes up): 160MB.
Pale Moon hovers around 278MB (though I will admit I have Adblock Latitude, Encrypted Web, and Greasemonkey installed, so it's not a fair comparison.) (The locking up behavior has been tested without add-ons, as well.)

So there appears to be very little room for subjectivity with regards to performance.

Currently at my school, so I cannot quite test a fresh profile; however, I have used busy boards on 4chan, and 8chan and used the catalog view. I do not experience the UI lock up (Intel Pentium 4 (hyper-threading disabled)). I do notice the scrolling is a tad bit sluggish; however, this is fixable in many other ways than switching to a multi-process structure.

The subjectivity comes from those who do heavy multi-tab, or multi-window browsing. I simply cannot run any browser when they are multi-process structured without causing my system to start using swap which causing everything to become sluggish.

I may have messed up on the terms here; I'm not entirely clear on the implications of multiprocess-vs-multithread. What I am saying is that whatever Chrome is doing is working, and working so well that Firefox is spending significant development time following suit.

Ah, understood. Just some background information, all browsers are multi-threaded (at least the main ones, including Pale Moon).

As a low-end hardware user (slow processor and limited RAM), I stand to benefit the most from these performance boosts.

It's purely anecdotal (I haven't run any hard scientific-type tests on this yet), but in my experience, Chromium on a Pentium 4 runs about as fast as Pale Moon on a Core 2 Duo.

Whatever Chrome is doing performs excellently, even on low-end hardware. What Pale Moon and older Firefox versions are doing, by comparison, is NOT running on low-end hardware very well. The only behavior "punishing those who cannot afford the latest hardware" is that which Moonchild is currently planning to proceed in.

I understand if supporting hardware regarded as obsolete is not a project goal, however.

You do for your browsing habit; however, those performance boosts for you, completely destroy the possibility of tab heavy or window heavy browsing others do.

*I do not have time to reread this post, excuse any ill conformed thoughts if there are any. :-P

[gpower2]

I totally agree with Moonchild on the disadvantages of inter-process architecture. Each process needs to re-initialize a lot of common code, or use a lot of inter-process calls to get it from another common process. This procedure is quite stressful on the CPU and RAM resources. However, when using a small number of tabs in Chromium browsers, one can without doubt experience a "snappier" browsing performance, since each tab is almost totally independent from one another.

That doesn't mean that the multi-process approach is the correct one. It may seem nice at first but after more tabs being loaded, you can witness a total PC degrading performance. It's almost as if opening different browser instances for each tab, like we used to before Firefox first came out.

Since nowadays most of the CPUs incorporate more than one CPU cores, parallelization of the code is a big must. However, not all code can be implemented in a true parallel way, and simply using OpenMP or other pragmas in the code doesn't make it parallel. Yes, they can help in a lot of ways, but you won't see the performance upgrade that you would expect.
I believe that one way to go, is to make each tab a true separate thread, under a common process in order to share common memory and reduce the inter-process calls. Perhaps a plugin-container thread could also be implemented, instead of a separate process as it is now (plugin-container.exe). That way, each tab could be executed from a separate CPU core, thus taking advantage of the multi-core CPUs. Of course that won't be the case in a real life scenario, however I believe that it could be a really nice approach, since, instead of trying to parallelizing common code, we would have a nice multi-threaded design which is parallel by itself.

I don't know what the current source code design is, and I would be really interested in it, if Moonchild could give a little more information.

[Moonchild]

Since nowadays most of the CPUs incorporate more than one CPU cores, parallelization of the code is a big must. However, not all code can be implemented in a true parallel way, and simply using OpenMP or other pragmas in the code doesn't make it parallel. Yes, they can help in a lot of ways, but you won't see the performance upgrade that you would expect.

By far most code can't be executed in a parallel way. It takes designing functions and procedures specifically with parallelism in mind. So yes, using carefully-placed pragmas to instruct/hint a compiler to parallelize code will help in some ways but it won't magically be able to re-write code into parallel-ready code.

I believe that one way to go, is to make each tab a true separate thread

It would be if it was a straightforward application with just passive and/or predefined content; but a browser doesn't work that way. Content, by design, is event-driven; that is simply how JS and dynamic HTML works and how the browser must deal with its content to remain performant. If you approach this with a classic (processor-)threading model then things are going to get much worse, especially considering you'll have a lot of things constantly calling into the common components while having no actual tasks to perform.

Perhaps a plugin-container thread could also be implemented, instead of a separate process as it is now (plugin-container.exe).

This is a separate process for 2 main reasons, neither of them are performance related and both of them are best left to the OS to handle:

• Crash-guard: If a plugin crashes, it will not take the rest of the browser with it. (stability consideration)
• Integrity/system privileges: The separate process is run with low integrity to prevent code that is out of our control from getting access to system areas. It has a lower privilege level than the main browser. (security consideration)

Also, this wouldn't be a prerequisite for a multi-threaded application, since IPC with plugins is fine either way and you don't want to duplicate plugin instances for each tab.

[Moonraker]

ww.ghacks.net/2015/11/27/firefox-electrolysis-multi-process-wont-come-out-this-year/w

Seems it is postponed again.

[LimboSlam]

Yep, until Firefox v45 at the very least. Seems to be a lot of time consuming e10's bugs still standing.

[Moonchild]

Seem they are trying very hard to get asynchronous panning and zooming to work on desktop which isn't working very well either (I personally had to file a few bugs on that after some casual runs of Nightly^[1][2][3]) which seems to be something important they want for e10s.

[1] bug #1227799
[2] bug #1227971
[3] bug #1228133 (actually just got solved)