Ubuntu Unity Proves Very Slow To KDE, GNOME, Xfce, LXDE

[bug77]

It can explain some of it, but not all of it. There's also the (hopefully just CPU) overhead of the compositing manager itself.

Shouldn't Kwin determine what is has to work on as soon as possible, in order to avoid doing unnecessary work, thus keeping the CPU overhead at a minimum?
I'm thinking of something resembling the z-buffer functionality: check early what you have to do and what you don't.
I don't know, maybe it's just me, but having to suspend compositing instead of the compositor being able to tell it doesn't have much to do in the first place seems to point at some implementation issues. And I don't necessarily mean that in a bad way, it's quite possible developers just didn't have the time to overhaul/tune that area yet.

[kraftman]

I don't know if there were previous benchmarks with fullscreen effect suspend both on and off(I don't remember any), but as far as I know compositing is normal to affect performance since it requires extra work from the hardware(actually I think there's comments about this and how it works in this thread here too ..!).

It does affect performance, but not with every driver. I'm running fglrx and there's no difference. With Open Source drivers and nVidia binary driver you should see a noticeable slow down with compositions being enabled. Some Phoronix benchmarks are so moronic, because they don't really benchmark driver speed (the best examples are Linux vs OS X and Intel hardware). I bet Kubuntu with suspended compositions turned on will simply kill crAppple OS.

[mrugiero]

What are the other official desktops of Ubuntu?
I know kubuntu was community, and I thought lubuntu and xubuntu were as well.
It certainly seems like Ubuntu is trying to make itself synonymous with Unity.

BTW, does it seem strange that kwin is faster than lubuntu? I don't recall the Wm they use but I'm fairly certain that it doesn't support composition.
Similarly, it looks as if mutter isnt always correctly detecting fullscreen windows and then unredirecting.

The only official desktop is Unity. But Ubuntu is not just the desktop. A slow boot time might be caused by a lot of things. Is the desktop usually the most relevant (individual) component, in terms of boot time? Likely, yes. But not the only one. I recall after upgrading to 12.04 I had a boot time larger than 1 minute, IIRC it was about 3 minutes, but it was due to a change in the interfaces defaults, nothing related to the desktop.

Yes, it seems strange to me the fact you point out. They use Openbox by default, and it doesn't support composition in any stable releases yet. There was some work in git in a GSoC, but I don't know how far it got. I believe it's merged to trunk. I think there might be place for optimizations there.
EDIT: WHOOPS, it seems I misrecalled about Openbox, the one which received work for compositing was Fluxbox. It is Openbox the LXDE's default, though, and it doesn't support compositing.

Ok, that somewhat resembles double buffering (except that this time it's per window). But if the WM only has to draw one "window" and then copy it to screen, is that enough to explain the rather sizeable gap in Michael's graphs?

There are a lot more operations going through with effects. As a start, something KWin did yesterday or so, making the translucency effect operate only on window movement might not be done in some of the other compositors, and thus wasting a lot more cycles with compositing on, and similar might happen with overlapped windows.
Also, even when it's not copyed to screen, I'm ALMOST sure the content of the windows gets updated while compositing is on, even when not showing, inside its buffers. So, even when overlapped by another window, it takes cycles.

Um, maybe this is naive but why wouldnt it always un-redirect when offscreen FB=screen size?

Then a really big window, which might be partially off the screen and still be a window, would always turn your compositing off.

I don't know if there were previous benchmarks with fullscreen effect suspend both on and off(I don't remember any), but as far as I know compositing is normal to affect performance since it requires extra work from the hardware(actually I think there's comments about this and how it works in this thread here too ..!).

Yes, it's being discussed

[liam]

Ubuntu and Unity aren't the same thing.

What are the other official desktops of Ubuntu?
I know kubuntu was community, and I thought lubuntu and xubuntu were as well.
It certainly seems like Ubuntu is trying to make itself synonymous with Unity.

BTW, does it seem strange that kwin is faster than lubuntu? I don't recall the Wm they use but I'm fairly certain that it doesn't support composition.
Similarly, it looks as if mutter isnt always correctly detecting fullscreen windows and then unredirecting.

[Rigaldo]

Interesting results, especially KWin fullscreen compositing vs fullscreen suspended. Compare this with the Phoronix results published earlier showing that with the Catalyst drivers give the same performance, fullscreen compositing enabled or not. What does this say about the Catalyst and Intel drivers? I guess performance should be the same whether or not compositing is enabled, non?

I don't know if there were previous benchmarks with fullscreen effect suspend both on and off(I don't remember any), but as far as I know compositing is normal to affect performance since it requires extra work from the hardware(actually I think there's comments about this and how it works in this thread here too ..!).

[bachinchi]

Kwin FTW

I love the screencap of the new Ubuntu login session scrolling right off the screen. I'm serious when I say: I don't think they test this stuff.

So can we conclude that Cinnamon would be similar to GNOME Shell in terms of performance?

I would say yes. But since Cinnamon uses a fork of Mutter (Muffin) ne can't be completely sure about it.

I'm eager to know what happens when using other hardware (and driver) combinations.

[liam]

Normally, a window writes into an offscreen framebuffer of the size of the window, which is later copied onto the screen at the position of the window. If the compositor can properly detect a fullscreen window, it can skip the copying and render to the screen directly. It's only the copying that makes the difference.

Um, maybe this is naive but why wouldnt it always un-redirect when offscreen FB=screen size?

[marek]

Ok, that somewhat resembles double buffering (except that this time it's per window). But if the WM only has to draw one "window" and then copy it to screen, is that enough to explain the rather sizeable gap in Michael's graphs?

It can explain some of it, but not all of it. There's also the (hopefully just CPU) overhead of the compositing manager itself.

[Hamish Wilson]

That's not how it works. Only the full screen window and windows beneath it lose the composition and it doesn't seem to affect the other screen at all. I'm doing that all the time because the only way I can get tear free video on my external monior is by using MPlayer without compositing with VDPAU output . I'm not sure how it affects performance though.

That is how it has been working for me with the Xfce compositor and suspending full screen windows, although since I have my dual head setup using Zaphod Mode so that I do not need to be bothered when a game changes my screen resolution, each monitor is treated as a separate screen in my case.

As much as I don't want to believe it. Unity is by far slower than Windows. It also takes forever to boot. There seems to be something hanging between login and logout that takes about 60 seconds to get passed it. The unity panel itself lags when scrolling up and down the menus and the dash menu has a big delay when opening. It feels like Windows 7 when you got infected with a bunch of malware 0_o

Yes, nobody said that you can not make a crap desktop on either Windows or Linux. Microsoft got Aero to work in the end - Ubuntu still has not gotten Unity to.

[molecule-eye]

Unity's no good

Interesting results, especially KWin fullscreen compositing vs fullscreen suspended. Compare this with the Phoronix results published earlier showing that with the Catalyst drivers give the same performance, fullscreen compositing enabled or not. What does this say about the Catalyst and Intel drivers? I guess performance should be the same whether or not compositing is enabled, non?