Making A Easy-To-Setup $50 Linux Multi-Seat Computer

[PeterKraus]

Interesting. I wanted to get a multiseat solution a year ago, when I used to have my office sharing a wall with the living room. I would have just used standard VGA connection and maybe an USB hub for the keyboard / mouse.

Though, considering the points some people made in this thread, the USB 2.0 might not be the right choice. I was wondering if Firewire (with DMA) or USB3.0 wouldn't have been better.

On the other hand, I would be interested in seeing that software part. Configuring (persistently) the new seat seems to be the biggest issue with the current state of multiseat, and also you need a graphics card head for each seat - that's where that onboard GPU comes in handy. Then, the only limitation becomes the length of your VGA/DVI cable....

Here's a great howto, by the way: https://wiki.archlinux.org/index.php/Xorg_multiseat

[frantaylor]

The lack of graphics acceleration in thin-client-like multiseat solutions has always been a put-off. .

SINCE WHEN do call-center operators need graphics acceleration on their workstations? These setups are GREAT for petition drives etc.

WHO NEEDS ACCELERATED GRAPHICS FOR DATA ENTRY? OMFG! 15 years ago we did this with monochrome Sun3 displays AND THEY WERE OVERKILL.

Hey why not some MORE hyperbole! Oh my God how can you POSSIBLY sit in front of a display with less than a zillion terabits of display bandwidth! I mean REALLY it's just IMPOSSIBLE to even LOOK at a screen if it can't update EVERY pixel for EVERY frame!

Really, people who program computers can blithely stick "ALWAYS" into their communications with humans and they don't expect to get it thrown back at them?

[schmidtbag]

SINCE WHEN do call-center operators need graphics acceleration on their workstations? These setups are GREAT for petition drives etc.

WHO NEEDS ACCELERATED GRAPHICS FOR DATA ENTRY? OMFG! 15 years ago we did this with monochrome Sun3 displays AND THEY WERE OVERKILL.

Hey why not some MORE hyperbole! Oh my God how can you POSSIBLY sit in front of a display with less than a zillion terabits of display bandwidth! I mean REALLY it's just IMPOSSIBLE to even LOOK at a screen if it can't update EVERY pixel for EVERY frame!

lol looks like somebody woke up on the wrong side of the bed this morning, and that side of the bed happened to have a concentrated energy drink.

But anyways, I see what you mean but for some cases hardware acceleration is needed. Yes, if all you are doing is typing a document then you don't need anything better, but if you are doing something like web browsing (which places like schools and libraries would often have) then no hardware acceleration takes a BIG hit, big enough to the point that other users would be directly affected.


As for anybody using a multi-seat setup where most of the seats aren't accelerated, I'd recommend looking into VirtualGL, which allows you to port GPU processing power over a network, which in turn allows you to get hardware acceleration on an otherwise unsupported platform. The nifty thing about it is you could use a real powerhouse of a GPU on a crappy netbook, or a virtual machine.

[bernieplugable]

I would like to see if USB 3 would improve this. I can see it being really useful, even for just families.

We do think families will like this kind of solution (assuming the kids are happy in Linux or the browser). I've run (an older USB multiseat solution) my own house for months at a time, and it's great to have only one computer to maintain. The only tension is all the old Windows game CDs we have laying around that Wine wouldn't run.

In terms of USB, Michael mentions the key problem -- but even on top of that, there are practical issues that mean USB 2.0 is a better overall solution in almost all cases today. USB 3.0 devices and infrastructure cost more. But the biggest issues are USB 3.0 cable length and stability. USB 2.0 is great in that there are cheap cable available providing long runs with no errors (we sell a 10 meter active cable like this). Whereas USB 3.0 is dodgy even beyond 1 meter. Also, it's great plugging lots of thin clients into a single 7 or 10 port USB hub. With USB 3.0, only 4 port hubs are out now (and the quality of them varies greatly). You want a setup like this to be rock solid (and with USB 2.0, it can be with good hardware).

So we expect USB 3.0 solutions to make their way onto the market, but actually USB 2.0 has a whole bunch of benefits (even given the constrained throughput). Remember that only the pixels that are changing are going over the bus (and compressed, at that). So it's actually surprising how performant the USB 2.0 solutions are.

Hopefully Michael will have the chance to do some more benchmarks and videos in his own lab, showing the full performance story. It's not for gaming or 1080p motion video, but it's great for information work and good enough for Youtube-quality video. That's not for everyone, but it meets many needs, especially at schools and non-profits.

I hope we get a few people here interested enough, to give it a try and see it in action. Thanks for posting!

[bernieplugable]

For those interested in some of the plumbing, Lennart Poettering (Red Hat) just put up a great post with details of what they've done in Fedora 17:

http://0pointer.de/blog/projects/multi-seat

[leif81]

This may be obvious to everyone else, but with the mass of wires it wasn't clear to me. Particular with specifically how the Plugable hub is connected to the host PC. I presume this is what's happening?

(PC)----USB---(Plugable hub)----DVI---monitor
   |                       |----USB---keyboard
   |                       `----USB---mouse
   |
   |----USB---(Plugable hub)----DVI---monitor
   |                       |----USB---keyboard
   |                       `----USB---mouse
   |
   `...

If that's the case, can we get one where the link between the host PC and the Plugable hub is wired RJ45 or wireless?

[chithanh]

Really, people who program computers can blithely stick "ALWAYS" into their communications with humans and they don't expect to get it thrown back at them?

Relax. I didn't say the lack of acceleration made them unsuitable for all purposes. Linux Multiseat¹ as we know it today started with the (now defunct) backstreet-ruby kernel patches and ever since, acceleration has been a problem. Since Compiz release in 2006, OpenGL composited desktops have become desirable. So I'd say it has become a major put-off in 2006, and was a minor put-off (but one nonetheless) before that.

---
¹ the whole article and this forum thread is only about Linux

[kebabbert]

This is similar to "thin clients" but different. This solution are called "zero state clients" or "ultra thin clients".

I dont like ordinary normal "thin clients" because they typically have a very weak CPU, and 256 MB RAM, and they have an OS that you need to patch and protect. You can upgrade with more RAM in the thin client. Thin clients are just like a very weak PC, you need to maintain and patch them. For instance, HP sells lot of thin clients.

This solution is instead something different and called "zero state clients", I really like this solution. Everything is run on the server and there is no cpu in the client that runs any software. The clients dont run any software and you can not upgrade them. If you google, there are several vendors that sell "zero state clients". Avoid "thin clients" - they suck. Imagine a PC driving several other workstations. Each client uses 2.5 Watt which is very good. Say your PC uses 100 watt, and say you have ten users, then each user will spend 10 watt. That is better than each user using 100 watt pc each = 1000 watt. However, the problem with "zero state clients" has always been graphics. 1280x1024 which this solution utilize, is not really usable.




Of course, Sun was first and introduced this idea of "zero state clients" some 25-30 years ago with the first SunRay client. And now others are copying the concept. Here is a research paper on this:
http://labs.oracle.com/features/teny...s/Nrthcutt.pdf

Now Oracle/Sun has released this software for Linux, so it installs under Linux too. "SunRay Server Software" can be downloaded from www.oracle.com. But you also need a client, for instance the Sunray2 client, it uses 4 watt, but goes up to 1600x1200, it weighs 0.38kg. Yesterday it was sold on Ebay for $20 a piece. Just plug it in into your router with a network cable and within seconds the login screen pops up. SunRays are heavily used in Enterprise environments (for instance, Microsoft use SunRay clients with Windows). For instance Sun had 19.000 SunRay clients worldwide, and only 38 sysadmins for the SunRay servers. Imagine how many sysadmins would be required to manage 19.000 PCs. It is much easier to manage a few servers, and then connect to the Servers via WAN, over internet. Some servers were in USA, and the SunRay users in India.

Oracle/Sun has just recently released a zero state client called SunRay 3+, which can drive two 30" screens with up to 2560x1600 resolution, simultaneously. That is certainly usable. My Solaris PC at home is driving a SunRay client, which my girl friend uses. According to old studies from Sun, a Sunray client typically uses 700MHz of cpu power when doing heavy Office usage. Thus, a quad cpu at 2.8 GHz can drive 2.8 / 0.7 x 4 = 16 office users. Oracle is now working to get lag free graphics with the SunRay clients, they will be able to play video in 1080p.
http://en.wikipedia.org/wiki/Sun_Ray

There is also a software SunRay client, that I have installed on my iPad. From my iPad, I can login to my Solaris pc and fire up Windows in VirtualBox, and watch flash movies on the iPad.

[perpetualrabbit]

I have some multi-seat systems in the office. I don't need slow displaylink systems, each seat has its own graphics card. It would be even better if each display output could have its own X server, but that is not possible right now for current nvidia and ati drivers, at least the proprietary ones. And with the open source drivers I lose so much performance and important features so they are no option. But I read something about open source drivers adding the possibility of multiple X servers per card.

But the biggest problem is per-seat sound and usb ports.

Now, in principle this should be an easy problem. Suppose you limit yourself to usb-audio headsets only.
USB is a tree structured bus. So, you could designate all usb connections originating from one particular usb hub for one seat.
So each seat has its own usb hub, either in the screen, or a separate hub. Anything you plug into that hub goes to that seat, be it a usb headset, or usb storage devices, whatever.

Anyone interested in investigating this with me? See what is available in consolekit, udev, gdm, xorg.conf etc. and what needs to be added to make this a reality?

[pingufunkybeat]

And with the open source drivers I lose so much performance and important features so they are no option.

I honestly cannot imagine what you lose in terms of performance and important features with open drivers, to the point where they are not an option IN AN OFFICE.

Do you do CAD and 3D modelling in your office? Or watch high-bitrate BluRays? Which features are missing?

Free drivers are a perfect match for office work. In fact, I find them to be superior to the blobs for regular desktop usage.


Anyway, this is a very interesting development. It's a shame that it's GNOME-only so far, but I'm sure that KDM will pick it up soon, as it has traditionally been much better with multi-seat than GDM. It will take a while for all this to trickle down to other distributions, but it shows how much important work the Fedora/RedHat people do.

I even forgive them for PulseAudio :P