Drafting Plans For X12, The X11 Successor

[Cyberax]

Network transparency lets me run some arbitrary application on a remote system, and have the display on my current desktop. Typically stuff like VNC and RDP don't export one application window, they export an entire desktop.

That can be solved in RDP and in VNC which support exporting of single windows just fine. I've personally used http://metavnc.sourceforge.net/ before.

In near future it might work like this:

1. You log on to remote computer with Wayland forwarding available.
2. You start wterm (Wayland terminal) which starts a remote daemonic Wayland server which might use local video acceleration (may be based on LLVMPipe).
3. The remote Wayland server communicates with your server over some sort of remoting protocol, maybe adapted VNC.
4. If you disconnect then the remote clients die and disconnect from their Wayland server which in turn dies if there are no more clients.
5. You can leave GUI application running in screen/tmux session. This way the remote Wayland server would just continue to working locally and you would simply re-establish connection once you reconnect.

There might be some issues with client-side decorations, but I think they are definitely solvable.

Another cool feature - it might be possible to migrate Wayland server from SSH session to the real local console.

[phred14]

Network transparency is a neat feature, but what does it actually provide to the user that something like RDP or VNC doesn't? You can call those gross hacks if you want, but the bottom line is that all the actual X developers have said that it's a poor architecture and that RDP is far superior. I trust that they know what they are talking about far more than some random posters on Phoronix, unless you can actually articulate what it is that we lose.

Network transparency lets me run some arbitrary application on a remote system, and have the display on my current desktop. Typically stuff like VNC and RDP don't export one application window, they export an entire desktop. So you're stuck with a slightly schizoid desktop-here vs desktop-there situation. You can work around it, but it's kind of ugly. I tend to use x11vnc to grab my existing work session from home, and that can be especially bad when the work display is better than my home display. You end up either scrolling a viewport or scaling, neither of which is very good. Also, at work I really only want to run one desktop environment - I don't want to start all of that crap for every system I use, nor do I want to wade through multiple desktop environments.

Practically all of us use this every blasted day at work - getting resource from elsewhere. Plus text-only doesn't hack it, because eventually some sort of graphical display becomes necessary for what we do here. (VLSI CAD) In fact, the "strategic direction," which I personally disagree with, is to put laptops on everyone's desk with a proprietary Windows-based X solution, and you get your engineering Linux access "from the local cloud." For my own purposes, I found that the laptop I was able to spec for the job was so beefy that I got some extra DRAM and just run Linux on it. I go remote for stuff that's beyond what the laptop can handle, or when there are back-level OS requirements that simply won't run on my laptop. (It's also several revs ahead of the "standard platform.")

This is the working world. Programming is part of my job, but it all has to do with delivering silicon, not with GUI programs. I need to grab resources where I can grab them, and there are often unique requirements for different tasks, so you grab this resource for this task and that resource for that task. I'm paid to get my job done, not to indulge in platform purity or stuff like that. I understand that X11 is old and crufty, and in fact I was an early adopter of xcb, including sending error reports along to software vendors when the locking problems became apparent. I don't care if it's X11, xcb, X12, or Wayland with something on top. I need some form of network transparency, and I need it to run arbitrary applications, not just those designed for it.

I have work to do! Which is what really annoys me when people can't look past the end of their own noses and use logic like, "I don't need network transparency, so it's not necessary at all!" or "That's so quaint!"

[Cyberax]

Well, I just would like to add some information about a solution to have the same functionnality than citrix or windows rdp on Linux.
There is a projet called "x2go" which does exactly the same as citrix but on Linux. It uses NX proxy for display redirection and it works very well...
So YES, there is a Linux solution and it's not a nighmare to install or use, and it's FREE goddamit...so try it, you will be surprised of the quality and the speed...

And it uses almost none of the famed 'network transparency' features of X instead replicating essentially what Wayland is going to do for remote clients.

[sbodmer]

The NT goons at work use Citrix virtualized desktops that provide remote GUI desktops to datacenters that are housed well over 1/3 of the world away. That is, quite literally, we have WAN connections on entirely different continents that provide Windows desktops remotely. I couldn't even dream of doing that with X11. It's a total friggen nightmare in comparison.

Well, I just would like to add some information about a solution to have the same functionnality than citrix or windows rdp on Linux.
There is a projet called "x2go" which does exactly the same as citrix but on Linux. It uses NX proxy for display redirection and it works very well...
So YES, there is a Linux solution and it's not a nighmare to install or use, and it's FREE goddamit...so try it, you will be surprised of the quality and the speed...

[Azpegath]

James Gosling argues, on a paper from 10 years ago, against network transparency. He talks about how he would re-invent a modern graphics system and his thoughts match 100% the Wayland design. Basically this is just common sense and I find odd how long the current flawed design has survived trough the years. Other than unix most other systems have had much better graphics infrastructure, i.e. leaner and meaner.
http://hack.org/mc/texts/gosling-wsd.pdf

Thanks for this, I found it very interesting. Well written and I think it's very noteworthy that the article was written 10 years ago, when GPU:s were nothing compared to now. He also proposes solutions to the whole "X11 compability" issue.

I also like how he wraps it up: "This window system design isn’t particularly radical: it’s more just pointing out that this is the way that X is going already, given the increasing predominance of application-side rendering libraries. Once you accept that fact and admit that it’s actually the right way to go, the design falls out, simply by stripping away legacy stuff that isn’t needed any more."

[Azpegath]

I'm a realist who doesn't think total rewrites with major semantic changes are the way to go with complex infrastructure software.

+1. Complete rewrites are cool and can be relevant when there has been a generation change in how things work (TM), which I guess can be said about the graphics stack for the desktop. But it takes time and it doesn't hurt to invest time in evolutionary improvements of what actually works.

[BlackStar]

That said, Wayland won't be efficient either. It seems to me that the best we could do with Wayland is buffer-scraping, which is about the least efficient way to do remote desktop. To make RD possible on Wayland would require that the client libraries be hookable and send their drawing requests to some sort of proxy that then sends the graphics stream over the network in an efficient way. Does the architecture allow for this? If not, then we're better off fixing up the X protocol and tweaking XCB than moving over to Wayland.

Of course the architecture allows for this. Wayland is completely flexible in this regard: it doesn't render anything by itself! In the Wayland world, it is the clients - not the server - who are responsible for rendering. Unlike X, the clients can actually use an efficient rendering library with network transparency and, again unlike X, reap the improvements as newer and more efficient protocols evolve. (Of course, you can also use X network transparency - Wayland *is* completely flexible in this regard).

Not only is this approach more efficient, it is also more logical. Wayland is a display server and a compositor and, as such, it composes and displays. It's not a renderer, nor a sound server and definitely not a network server.

Remember the Unix philosophy? "Write programs that do one thing and do it well." Too bad that most people seem to forget it these days.

[drag]

I took a look at the RDP spec and it appears that it tunnels GDI commands through the protocol, but since compression and caching are part of the protocol, it can do a better job of utilizing bandwidth than X can.

Bandwidth is the least of X11's problems. People have come up with schemes for compressing stuff on X11, but it only usually makes things worse.

Why?

Because latency is the problem, not bandwidth. No amount of compression is ever going to help it. The protocol is just obsolete and the applications that use it are mostly very poorly designed.

X11 is just shitty and outdated. It was pretty cool, but the world is moved on. Even though RDP is superior for desktop applications then X11, RDP itself has been outclassed massively by competitors.

The NT goons at work use Citrix virtualized desktops that provide remote GUI desktops to datacenters that are housed well over 1/3 of the world away. That is, quite literally, we have WAN connections on entirely different continents that provide Windows desktops remotely. I couldn't even dream of doing that with X11. It's a total friggen nightmare in comparison.

People who cheerlead X11 as some sort of remote GUI feature are so far out of touch of reality it's not even funny. Remote desktop and remote GUI applications are now mainstream. People regularly use 'Goto Meeting', Citrix virtual desktops, and all sorts of other stuff. People work from home or on business trips using remote GUI applications all the time. NOT A SINGLE ONE OF THEM USES X11.

I can run X11 on Windows. I can run X11 on OS X. They work perfectly fine. Just as well as they do on Linux. But you know why nobody actually uses it besides some sys admins? Because it sucks and there are far better remote GUI solutions out there.

If X12 can solve this then this would be fantastic. But in terms of network remote GUI access Linux is at least 10 years behind Microsoft. And Microsoft is years behind Citrix. If I am working on a document and my network drops out then I lose the application and what I was working on... I wouldn't have to put up with that shit if I was using Windows. I just reconnect and the application would still be running just fine. That is, very simply, the reality and no amount of compression is ever going to fix fundamentally obsolete designs.

[siride]

Network transparency is a neat feature, but what does it actually provide to the user that something like RDP or VNC doesn't? You can call those gross hacks if you want, but the bottom line is that all the actual X developers have said that it's a poor architecture and that RDP is far superior. I trust that they know what they are talking about far more than some random posters on Phoronix, unless you can actually articulate what it is that we lose.

I took a look at the RDP spec and it appears that it tunnels GDI commands through the protocol, but since compression and caching are part of the protocol, it can do a better job of utilizing bandwidth than X can. The thing is, while theoretically, there should be no difference between a local window and a remote window, the fact of the matter is that the protocol, any plain old graphics protocol, really, is too bandwidth and latency inefficient to work over slow network connections. RDP uses smart compression techniques and a lot of caching on the client side to reduce bandwidth. GDI primitives, like pens and brushes, can be saved on the client and then only need be referenced by cache index for future operations with them. There's no point in putting that kind of logic in a local connection because it just wastes CPU and memory. It's faster to just send these things over the local socket than it is to play around with caches and compression.

X could win if it used a slightly different protocol for non-local connections. I don't mean in terms of TCP vs. unix domain sockets, I mean that if a client detected that the server it was talking to was on another machine, it would encode the X requests differently (with compression) and make use of caching (the server would have to play along here). XCB would make this a lot more feasible, but XCB has seen little uptake outside of being used as a backend for libX11 and mesa. Regardless, the X protocol as it is just isn't efficient for remote display.

That said, Wayland won't be efficient either. It seems to me that the best we could do with Wayland is buffer-scraping, which is about the least efficient way to do remote desktop. To make RD possible on Wayland would require that the client libraries be hookable and send their drawing requests to some sort of proxy that then sends the graphics stream over the network in an efficient way. Does the architecture allow for this? If not, then we're better off fixing up the X protocol and tweaking XCB than moving over to Wayland.

[Cyberax]

Excellent point. Those of us who have become accustomed to network transparency and who benefit from it always have difficulty explaining it to people who have never used it and who don't understand its value.

Well, some of us use network transparency with Windows and RDP. And let me tell you, it blows X out of water.

It's great to run a GUI program on a server from my notebook, then turn off the notebook, go home and then reattach to the same session and continue as if nothing has happened. That's something that X hasn't ever been able to do without hacks (like a proxy display server).