Legacy X and network transparency

Posted Nov 5, 2010 5:52 UTC (Fri) by PO8 (guest, #41661)
Parent article: Shuttleworth: Unity on Wayland

Folks should keep in mind that even with Wayland world domination in place X won't be going anywhere anytime soon. The plan is to provide a legacy X server hosted by Wayland. This will enable both remote and local existing X clients to work just as they always did on the local display. The only thing that won't work over the wire is new Wayland-only apps. This will be sad, but given the nature of graphics use by these apps their local-ness won't be so evitable anyhow.

Remoting a window manager, in particular, hasn't been a terribly useful option for a long time.

Legacy X and network transparency

Posted Nov 5, 2010 7:13 UTC (Fri) by jzbiciak (guest, #5246) [Link] (16 responses)

Reading the 20,000ft version of how Wayland works, it seems like it ought to be possible (eventually) to run a Wayland app over the network also. The main difference is that it looks like you'd use a simple bitmap-oriented VNC-like protocol between the Wayland client and the compositor rather than the complex beast that X is. (I don't know if anyone has proposed this--it just looks to me like the obvious way to do it.)

I'm actually quite OK with this, since it plays well into advances we've made (fast CPUs == fast compression, and we have ever increasing bandwidth), and does a better job of tolerating the one major bit that hasn't advanced much: round-trip latency.

I have an X application I sometimes need to run remotely over a VPN link over VDSL. I have gobs of bandwidth, but the RTT sucks. The app is barely usable. In contrast, Windows Remote Desktop and VNC both work just fine over the same link. Both of the latter seem to be more of the "dumb bitmap plus compression" school of thought, and that seems to work pretty well with modern setups.

Legacy X and network transparency

Posted Nov 5, 2010 7:34 UTC (Fri) by dlang (guest, #313) [Link] (7 responses)

and this approach throws away the huge advantage that the Wayland folks (or at least their advocates in this thread) are claiming, the ability to take advantage of the powerful local graphics capibility.

Legacy X and network transparency

Posted Nov 5, 2010 12:50 UTC (Fri) by i3839 (guest, #31386) [Link] (5 responses)

That is not necessarily true. If you chop up the window into a bunch of tiny textures, you can cache it all on the graphics card and only update the bits that changed. So it should be still plenty fast, because there's only one round trip.

The main advantage of Wayland is that it simplifies the whole graphics stack enormously. It uses DRI2/KMS, just like X does, so it doesn't give extra possibilities or a magic performance increase.

Legacy X and network transparency

Posted Nov 5, 2010 15:19 UTC (Fri) by drag (guest, #31333) [Link] (4 responses)

YES THIS.

Wayland is much simpler because it depends on a modern graphic stack. It'll be faster then X, though, simply because it's much less overhead and cleaner implementation. It won't be magical, of course. Only modest improvements. Probably be better in terms of battery life....

There is also no reason why you need to give up X Windows to use Wayland. I use X Windows just fine in Microsoft Windows. Also lots of people use X Windows just fine in OS X. Given that Wayland is naturally composited interface then having a Wayland-specific X Server that draws to off-screen buffers will allow natural integration and backwards compatibility with current applications.

Not that there is a Wayland DDX like there is for MS Windows DDX and XQuartz DDX, but it's certainly going to be a requirement. It's one of those things that will have to be made before Wayland is usable.

Applications that use Wayland will immediately be able to benefit from being 'native wayland', but X apps won't get lost out in the cold.

Legacy X and network transparency

Posted Nov 6, 2010 4:22 UTC (Sat) by rqosa (subscriber, #24136) [Link] (3 responses)

> Applications that use Wayland will immediately be able to benefit from being 'native wayland'

These native Wayland apps use DRI2 to draw to offscreen buffers, right? So, isn't it true that there's no reason why X clients couldn't also be made to use DRI2 to draw to offscreen buffers?

And in that case, there's no significant speed penalty to using X (because the X clients then are doing direct rendering without needing to go through the X server (as in AIGLX)), right?

And if Wayland doesn't have a speed advantage over X, then what is its advantage?

Legacy X and network transparency

Posted Nov 6, 2010 6:04 UTC (Sat) by PO8 (guest, #41661) [Link] (2 responses)

The big advantage of Wayland is simplicity. Because it is so much simpler to implement than X, our tiny pool of X developers is better leveraged. Because modern applications typically just want to emit OpenGL at the end of the day (albeit maybe by some client-side library such as Cairo) and most modern hardware directly supports OpenGL, having X "get in the way" just ticks app developers off by making their job harder. At some point, X starts looking like a huge bag on the side of app interactions with the display, hence Wayland.

Legacy X and network transparency

Posted Nov 8, 2010 16:14 UTC (Mon) by renox (guest, #23785) [Link] (1 responses)

Given that apparently you still need to keep an X Server for legacy application AND the ability for application/toolkit to speak X when they want to have network transparency, adding Wayland will add code, not remove code..
So this 'simplicity' isn't very convincing: yes, Wayland itself is simple, but as it's not a complete solution, the result won't be simple!

Legacy X and network transparency

Posted Nov 9, 2010 0:19 UTC (Tue) by alankila (guest, #47141) [Link]

Well, hopefully a method to move a single application's window over network can be found, somehow. If Wayland is to win, it absolutely has to replace X, and that includes some kind of support for this feature. So you can be pretty sure that use of X will be seen as a bug if we actually do get Wayland-managed display system going.

Legacy X and network transparency

Posted Nov 5, 2010 17:50 UTC (Fri) by jzbiciak (guest, #5246) [Link]

If I'm not mistaken, it's still possible to use the video card for acceleration, even if what you generated didn't go to the display. Otherwise, it seems like rendering in the client and sending to the compositor wouldn't work.

You just lose the shared-memory efficiency when you hand over the results, since the client's video card and the compositor's video card are in two entirely different boxes.

Legacy X and network transparency

Posted Nov 5, 2010 9:07 UTC (Fri) by marcH (subscriber, #57642) [Link] (3 responses)

> ... and does a better job of tolerating the one major bit that hasn't advanced much: round-trip latency.

Err... are you seriously expecting network latency to become better? You know that it depends on the speed of light, right?

I know it could be better than Ethernet *on the LAN* but I doubt that X protocols are soooo chatty they would feel any difference.

> I have an X application I sometimes need to run remotely over a VPN link over VDSL.

DSL offers notoriously bad round trip times (20-30ms) because of the massive amount of Forward Error Correction. You should either look for an ISP that allows to tune your FEC (as explained here http://www.dslreports.com/faq/2182), or for an entirely different and better access technology like DOCSIS. Maybe even 3G has better latency than DSL. Anyone knows?

Legacy X and network transparency

Posted Nov 5, 2010 15:22 UTC (Fri) by centenary (guest, #71028) [Link] (2 responses)

> Err... are you seriously expecting network latency to become better? You know that it depends on the speed of light, right?

I think you're misreading his point. His point *is* the fact that network latencies won't improve (which you're also saying here). Since network latencies won't improve, bitmap-oriented protocols have an advantage since X-forwarding performs poorly under network latency.

Legacy X and network transparency

Posted Nov 5, 2010 17:58 UTC (Fri) by jzbiciak (guest, #5246) [Link] (1 responses)

Yeah, that was pretty much my point: Bandwidth has (and will likely) continue to make fairly large leaps, while latency will make incremental gains at best or occasionally go backwards depending on the technology you use.

I didn't call out physics as the cause. I figured the speed of light should be pretty obvious in this crowd. :-) As for using DSL vs. something else: At least I'm not using satellite. *shudder*

In the end it wasn't a technical decision on my part anyway: The state of broadband being what it is around here, my shopping experience for a provider amounted to telling the sales person "I run servers", and seeing what happened. The cable guys told me "have a nice day," whereas the DSL guys asked "with or without static IP?" It may've changed since then, but does it really matter? I'm now spiraling way off topic.

Legacy X and network transparency

Posted Nov 7, 2010 15:52 UTC (Sun) by marcH (subscriber, #57642) [Link]

> I'm now spiraling way off topic.

With your shopping experience yes maybe... on the other hand the latencies of broadband technologies is quite relevant.

Legacy X and network transparency

Posted Nov 5, 2010 15:48 UTC (Fri) by deepfire (guest, #26138) [Link] (3 responses)

> I have an X application I sometimes need to run remotely over a VPN link
> over VDSL. I have gobs of bandwidth, but the RTT sucks. The app is barely
> usable. In contrast, Windows Remote Desktop and VNC both work just fine
> over the same link. Both of the latter seem to be more of the "dumb bitmap
> plus compression" school of thought, and that seems to work pretty well
> with modern setups.

Well, this is somewhat of a strawman. The fact that X is not implemented particularly efficiently does not mean that X is not fundamentally better than dumb protocols.

If you want to compare apples to apples, see NX, the heavily-optimised implementation of the X protocol. I use it daily, and FWIW it beats crap out of the competition.

Legacy X and network transparency

Posted Nov 6, 2010 6:20 UTC (Sat) by butlerm (subscriber, #13312) [Link] (2 responses)

NX is orders of magnitude better than X over a WAN, but in my experience it still pales in comparison to RDP.

That said, I sure hope someone is looking at a mid-layer API that can be adapted to virtually any combination of user interface toolkit and display communication protocol without having to reduce everything to a bitmap first.

Why should nearly any kind of application be programmed to an API that is designed to be non-remotable? That is the highway to balkanization. What we need is a generic mid-layer API that can reasonably support both scenarios without the historical infelicities of X, so you do not have to re-port your entire application just because you want to run it remotely on occasion, preferably without feeling you are watching satellite tv during a snowstorm.

Legacy X and network transparency

Posted Nov 6, 2010 9:37 UTC (Sat) by quotemstr (subscriber, #45331) [Link]

Sorry. I'm a heavy user of both RDP and NX, and I definitely prefer the latter. Not only is it actually faster, but there's more fine-grained control over authentication, forwarding mechanisms, and so on with X than there is with RDP.

Legacy X and network transparency

Posted Nov 6, 2010 13:09 UTC (Sat) by deepfire (guest, #26138) [Link]

Ditto, my experience with RDP and its kind is what prompted me to switch to NX in the first place.

Legacy X and network transparency

Posted Nov 8, 2010 8:16 UTC (Mon) by nix (subscriber, #2304) [Link] (2 responses)

And because remoting a window manager is a bad idea (odd, I did it last week and it worked fine), we can just drop the whole idea of a window manager! Nobody needs that flexibility, right? Everyone can just use Metacity, perhaps with some extra themes! There are no competitors worth speaking of!

(oh, wait.)

Legacy X and network transparency

Posted Nov 12, 2010 3:07 UTC (Fri) by jmorris42 (guest, #2203) [Link] (1 responses)

> we can just drop the whole idea of a window manager!

Yea, that is just an extra helping of fail to go along with ditching network transparency. So not only are we supposed to be happy tossing "The Network is the Computer, the Computer is the Network" we also lose "Mechanism not policy" along with it.

After both of those are gone, might as well just buy a Mac and be done with it.

Ya know, one of the attractions of Free Software for me was the hope for freedom from being abandoned by a vendor. But with the lemming like action of the distributions chasing "The Year of Linux on the Desktop" it looks like we (we the *NIX loving folk who were the early adopters) are about to be abandoned. Thankfully we will at least have the option to fall back to a distribution preserving the *NIX way.... even if we have to fork it off from an existing one and maintain it. Right up until Firefox and Chromium go Wayland and drop X support, then things might get messy.

Legacy X and network transparency

Posted Nov 14, 2010 21:07 UTC (Sun) by nix (subscriber, #2304) [Link]

Even then, Konqueror should still work. I don't see Qt dropping X support: after all, they already support, what, three or four or is it five different graphics layers? Adding Wayland shouldn't require *them* to drop X :)