X.Org Server To See More API/ABI Breakage

[liam]

no it won't, it will work the same, but there is no magic performance boost,

Dave.

That's not what Keith Packard said. See the april lwn article (sorry I don't have the link).
The reasoning was that direct rendering would never deal with X (so little real change there) but X drawing would stay vigoursly in process requiring a context switch only once every vsync, and them it only hands off a handle rather the buffer content.
Not a huge improvement, but certainly not a performance regression.

[entropy]

That's not what Keith Packard said. See the april lwn article (sorry I don't have the link).

I guess you're referring to this one?

http://lwn.net/Articles/491509/

The performance of X typically becomes "abysmal" when it runs on top of another window system like Windows or Mac OS X because the server has to do a lot of bulk memory copies. But, because Wayland is cooperating with X, you get "full speed 2D rendering" while direct rendering is unchanged. Buffer swaps will happen every 16ms, when the X server tells xwayland it has new content and Weston tells the kernel to display it. Putting X atop Wayland will actually "reduce the context switches to once every 16ms", which may make it perform better than native X, Packard said. He doesn't have any numbers, and the effect will likely be imperceptible, but it could be faster. From the audience, Greg Kroah-Hartman also noted that it may result in power savings.

[liam]

I guess you're referring to this one?
http://lwn.net/Articles/491509/

That's it. Thanks!

[ninez]

That's not what Keith Packard said. See the april lwn article (sorry I don't have the link).
The reasoning was that direct rendering would never deal with X (so little real change there) but X drawing would stay vigoursly in process requiring a context switch only once every vsync, and them it only hands off a handle rather the buffer content.
Not a huge improvement, but certainly not a performance regression.

Keith never says in that Article that Wayland WILL have a performance boost. He says he has no numbers to back the claim, that it MAY be faster and that it will likely not even be noticeable. I don't know where you got the idea he claims otherwise. It sounds like a hypothesis or hunch and is very foggy at best.

EDIT: i also don't think anyone claimed there would be a regression. I did point out that for me having to switch from using Nvidia to Nouveau in order to use Wayland meant that any 'performance boost' really didn't mean anything because switching from nvidia 307.11 binary driver to using Nouveau is a HUGE performance regression. ~ which is true, but has nothing to do with Wayland itself.

[liam]

Keith never says in that Article that Wayland WILL have a performance boost. He says he has no numbers to back the claim, that it MAY be faster and that it will likely not even be noticeable. I don't know where you got the idea he claims otherwise. It sounds like a hypothesis or hunch and is very foggy at best.

Fair enough that he didn't say it would definitely improve performance, but the fact that Keith Packard even said it makes it more than a little likely to be the case. I mean, is there anyone who is more familiar with both X and Wayland than him?
I don't typically like appeals to authority, but people at the conference had ample chance to contradict him (and there were most definitely people there who would be in a better position than either of us to contradict him), so I take that to mean his reasoning was sound, and barring something unforeseen, the expectation of a small performance increase is existing in the minds of the experts.
All that aside, his reasoning makes sense. What part of that do you disagree with?

[ninez]

Fair enough that he didn't say it would definitely improve performance, but the fact that Keith Packard even said it makes it more than a little likely to be the case. I mean, is there anyone who is more familiar with both X and Wayland than him?
I don't typically like appeals to authority, but people at the conference had ample chance to contradict him (and there were most definitely people there who would be in a better position than either of us to contradict him), so I take that to mean his reasoning was sound, and barring something unforeseen, the expectation of a small performance increase is existing in the minds of the experts.
All that aside, his reasoning makes sense. What part of that do you disagree with?

I don't disagree with his reasoning... I disagreed with what you said to Dave Arlie (who is also someone i would consider an 'expert'). Everything Kieth said was *very ambiguous*, meaning he didn't actually say anything about a concrete performance boost. Furthermore, he said "the effect will likely be imperceptible, but it could be faster." So it will LIKELY be impercptible ~ that doesn't sound re-assuring, as far as any real performance boost. ..and the rest of that statement is speculative.

does that clear up what i meant?

[johnc]

One thing's for sure... when I get my Geforce 690 there's gonna be a performance boost... muaha

[liam]

I don't disagree with his reasoning... I disagreed with what you said to Dave Arlie (who is also someone i would consider an 'expert'). Everything Kieth said was *very ambiguous*, meaning he didn't actually say anything about a concrete performance boost. Furthermore, he said "the effect will likely be imperceptible, but it could be faster." So it will LIKELY be impercptible ~ that doesn't sound re-assuring, as far as any real performance boost. ..and the rest of that statement is speculative.

does that clear up what i meant?

Clearly we have a bit of a communication problem here
My problem with Dave's comment was his seeming dismissive tone with respect to Wayland, X, and performance improvement. While I certainly agree about it no being a magic bullet, I think such a statement doesn't have a lot meaning behind it in that display managers are complicated enough that I couldn't imagine one being both capable and easy enough to implement so as to qualify for magic bullet status. To counter his rather blunt assertion of something that is, at the very least, uncertain, I used KP's comment. While I have absolutely nothing but respect for Dave, his knowledge, and contributions, I think this particular area is one which KP is a bit more involved in, or, at a minimum, enough of an expert to cause a bit of doubt with regard to Dave's comment.
Now, I imagine the point Dave was making was that people shouldn't expect Wayland to magically speed up, say cairo, without subequent work being done elsewhere (like, in cairo). In short, a compositor/input handler can be a bottleneck, but it isn't necessarily, and in Linux's case isn't, the biggest one.
Lastly, since entropy posted the actual article, and I agreed that he got the right one, I'm not sure why we're having this disagreement. I don't feel like I misrepresented KP. What I said is technically true. KP, being paraphrased, thinks there could be a performance increase, and I gave the reasoning, which was consistent with what the quote says. The only sticking point is that I failed to specify that it wasn't a certainty, but I did specify that any improvements would likely be minor (whether or not that is a "real performance boost" doesn't matter).
Does that make sense?

[airlied]

Clearly we have a bit of a communication problem here
My problem with Dave's comment was his seeming dismissive tone with respect to Wayland, X, and performance improvement. While I certainly agree about it no being a magic bullet, I think such a statement doesn't have a lot meaning behind it in that display managers are complicated enough that I couldn't imagine one being both capable and easy enough to implement so as to qualify for magic bullet status. To counter his rather blunt assertion of something that is, at the very least, uncertain, I used KP's comment. While I have absolutely nothing but respect for Dave, his knowledge, and contributions, I think this particular area is one which KP is a bit more involved in, or, at a minimum, enough of an expert to cause a bit of doubt with regard to Dave's comment.
Now, I imagine the point Dave was making was that people shouldn't expect Wayland to magically speed up, say cairo, without subequent work being done elsewhere (like, in cairo). In short, a compositor/input handler can be a bottleneck, but it isn't necessarily, and in Linux's case isn't, the biggest one.
Lastly, since entropy posted the actual article, and I agreed that he got the right one, I'm not sure why we're having this disagreement. I don't feel like I misrepresented KP. What I said is technically true. KP, being paraphrased, thinks there could be a performance increase, and I gave the reasoning, which was consistent with what the quote says. The only sticking point is that I failed to specify that it wasn't a certainty, but I did specify that any improvements would likely be minor (whether or not that is a "real performance boost" doesn't matter).
Does that make sense?

It was a reply to uid313's comment not yours. He made a fanboi comment, and I corrected him.

If I paraphrased Keith as "An possibly imperceptible performance boost", seems like a contradiction in terms to me.

The thing you might have a couple context switches, but you might also have a few more introduced elsewhere like the input paths, the latency might go up as well as the throughput.

Keith was engaging in a bit of marketing really, since no measurements are possible at this point. I generally prefer my performance boosts to come with a measureable difference, and I don't mean gtkperf gets faster, since nobody has ever correlated that with a real world difference.

wayland won't magically speed-up X rendering, the same operations happen. In fact something like SNA would speed it up a lot more than wayland, since it actually tries to accelerate X rendering, where wayland does nothing different wrt X rendering really vs having a compositing manager running on X now.

Dave.

[airlied]

It was a reply to uid313's comment not yours. He made a fanboi comment, and I corrected him.

If I paraphrased Keith as "An possibly imperceptible performance boost", seems like a contradiction in terms to me.

The thing you might have a couple context switches, but you might also have a few more introduced elsewhere like the input paths, the latency might go up as well as the throughput.

Keith was engaging in a bit of marketing really, since no measurements are possible at this point. I generally prefer my performance boosts to come with a measureable difference, and I don't mean gtkperf gets faster, since nobody has ever correlated that with a real world difference.

wayland won't magically speed-up X rendering, the same operations happen. In fact something like SNA would speed it up a lot more than wayland, since it actually tries to accelerate X rendering, where wayland does nothing different wrt X rendering really vs having a compositing manager running on X now.

Dave.

But of course anyone who studied how wayland and X works would know this already.....

Dave.