Hacker News Reader

Comment by prodigal_erik | Hacker News Reader

I find it strange to call it 1.0 while remoting is still on the todo list. The age of vertical scaling, owning one big computer and sitting at its console, is well into decline.

View on HN

Wayland is essentially to effort to move responsibilities to different layers. Some things are pushed down to the kernel (memory management, command scheduling, mode setting) or similar infrastructure (OpenGL, udev). Some things are pushed upwards to the clients (font rendering, window decorations, vector drawing, remote access). This means Wayland has much less to do compared to X.org and that is a good thing, because it removes code duplication and unecessary complexity.

http://wayland.freedesktop.org/faq.html

Even with cloud computing, all the rendering and half of the other processing is done client-side. Pushing pixels or low-level drawing commands over the network is not the way technology is heading.

The current trend in scientific computing is to put GPUs on the server and push pixels to the client, e.g. http://www.theaustralian.com.au/australian-it/software-switc...

For which they're using VNC, not X forwarding. There are a lot of ways to push those pixels that don't involve X.

Oh, it's a trend? How often is scientific computing about pixels anyway? I'm willing to bet not very often. Most of the time the results are a bunch of digits, a few matrices at most - a configuration, a solution to a problem.

Whilst I don't want this to happen, what about Cloud Gaming? (http://en.wikipedia.org/wiki/Cloud_gaming)

In this case, the "client" would be the game render server. Your terminal -- which has the display -- is the "server."

This is already the way OnLive, for example, works. All the rendering is done on the server side, which corresponds to the Wayland/X11 client.

Cloud gaming can't work on the Internet as it currently exists. Latency is just too high and too variable, especially for consumer-grade connections when they get near saturation. When bufferbloat is a thing of the past, it might be possible.

However, even smartphones are coming with highly capable GPUs these days, so within a few years, any CPU no matter how cheap will be able to handle at least casual gaming at HD resolutions.

Having a release people can start using now to test their apps against and write new apps for is the most important next step for them.

Remoting isn't as important when you have options like VNC to fall back on. Its a minor inconvenience at the most and I believe the've made the right call to ship something usable now.

How could any unix user conceive that remoting is not important. 99% of the gui I launch are remote. In fact I do not even know where are half the workstation I use.

Regularly, I have to give support to portugal or italia, they open a internet connection, so that I can log on their system. With a simple ssh -Y, I can perform all the investigations I need.

Probably because most unix users just use console ssh or web GUIs for remote access.

I personally find vnc or NoMachine NX better than X sharing. The main reason for this is that my programs don't quit if my connection goes down. While I do hope for a good remoting system, I'm not sure that 'ssh -Y' is the best way to do it any more.

ssh works out of the box everywhere. I can be connected on as many hosts as I want, launch any graphical application, anywhere and they just display correctly. I can not imagine launching a different X server almost for each application I am running. I have a single X server and often tenth of clients running on three of four different hosts. And when I have to access an "external site", I just phone them, so that they "open" the connection, then I do my ssh -Y and I can launch many GUI running on all their hosts to investigate their issues.

I have NoMachine NX on my home computer. It is very useful, but it is far more heavy weight than a simple ssh with X11 forwarding.

What X needed was to slim down. That's something I can get behind. But the Wayland folks went too far, and the rationale seems to have been "because this is our first real opportunity to break from X"... in other words, because they could. What we needed was an X12, not something different for the sake of being different: evolution, not revolution.

1.0 means a different thing to developers than it does to consumers. API backwards-compatibility is a very important thing, and calling this 1.0 is a signal to developers that it's safe to start building things on top of Wayland that should keep running without modifications for a very long time.

Perhaps they figure VNC and SSH do enough of the job for most people. Certainly works for me... I can't remember the last time I needed more than ssh to work on some remote machine.

Did you use SSH with X forwarding though?

X over SSH in my experience is much harder to configure, and slower

Some people claim it depends on adequately configuring X, which may be true, but it's much easier to configure VNC over SSH to have an acceptable performance

I think the main argument for remote X is that it is so easy to use between 2 true X11 environments. You just `ssh -X` to your target and start running programs.

On Windows or OSX though... If you were going to install X11 why not skip it and just install VNC?

Maybe I don't want to run a full X session on my remote machine, just to run a simple X client and have it displayed on my local desktop?

VNC is no alternative for running individual applications remotely. Remote desktop != remote X.

That's exactly what 'ssh -X' is for.

(But tastes much better as 'ssh -XC')

True, for one app, ssh -X is usually better

Still, most of the time, running only one application is not the usage model. Or it's a simple thing that doesn't need X, just ssh

Remote desktop allows remoting only single applications, though. That's how stuff like the XP mode with Virtual PC on Windows 7 works.

X protocol remoting is almost never what you want; with the client / server relationship swapped, all[1] the apps that you're running die once you disconnect. If you assume unreliable networking - not rare at all in many remote work scenarios - it's a recipe for frustration.

[1] Yes, I know there are exceptions.

Please don't tell me what I "almost never want", considering it's a feature that I do use all the time. My LAN is quite reliable and fast, thanks very much.

+1 ... X protocol remoting is amazing.

Amazing until a connection times out and apps close without being resumable

On a LAN this doesn't happen unless the power goes down.

Doesn't happen with NX.

It's not that everybody almost never wants it, it's that almost nobody ever wants it.

Probably should have been "almost never what one wants," but that's quibbling over English. Even if it were true that there were no mechanism for doing remoting on Wayland (which isn't the case; remoting just moves to a separate layer in the stack, as handled by NX or VNC or whatever), it would still be a reasonable sacrifice, because it's not what most users want. Doing X remoting is the exception, not the rule, and there's no reason for the *nix world to get stuck in the Windows mentality of having to be backwards-compatible with every application and protocol written since the dawn of time. Having a huge networking layer shoved into the middle of the display stack has measurable costs in terms of both maintainability and performance, and users that have any use case other than yours suffer as a result. Android, which is now the most-installed Linux "distribution" (such as it is) ever, had to invent its own display layer because X was a terrifying, bloated hacktastrophe, and they come up with something specific to their platform such that now no other Linux users can benefit from any of their display work. That sucks. Keeping network transparency as a core feature doesn't pass the cost-benefit smell test.

Well, having spend all my day on an Xterminal, I think you are stuck in Windows mentality. Unix mentality is network transparency. Just inform you about plan9. In unix, you do not care where your program is running, it just runs and you can interact with it on your X server. VNC is windows mentality and it just happen to be useful on unix because X toolkit layers are network hogs.

I haven't owned or operated a Windows machine in almost a decade, and have ten terminal tabs open as we speak; pretty sure my Unix cred is intact, thanks. And I thought the Unix philosophy was to do one thing and do it well (Wikipedia agrees: http://en.wikipedia.org/wiki/Unix_way#McIlroy:_A_Quarter_Cen... ). Separating the thing that paints the screen from the thing that pushes pixels over the network seems in perfect keeping with this philosophy. Bash isn't network-transparent; if you want to use it over a network, you use SSH.

Bash is network transparent: you change directory in a nfs filesystem or in a curlftpfs the same way as on a local disk.

X separates the thing that paints the screen (the X server) from the things that computes (the X client).

Your idea or pushing pixels means that you need a virtual screen on the machine where programs runs and you want to copy this virtual screen to your screen. Do you imagine the complexity of window management with such a setup ? Have you already launched many VNC clients at the same time ?

XPRA seems quite cool, and not that well known, but it occasionally gets mentioned when Wayland Remoting comes up

http://xpra.org/

"Screen for X"

I kind of disagree. The entire point of Wayland is to remove deprecated, crufty stuff to make a new form of display server. Remoting (in its current form) was on the list of stuff to remove.

If you need remoting, for probably a fairly long way into the future, X11 will still be supported and you can use that. If you want the new shininess of Wayland, sorry, you have to live without remote capabilities.

Reply on news.ycombinator.com