| 1 |
[snip] |
| 2 |
>> If I throw out installing a separate OS on a separate machine for each |
| 3 |
>> workstation and all of the proprietary thin-client protocols, I think |
| 4 |
>> I have 3 options: |
| 5 |
>> |
| 6 |
>> 1. Connect monitors, USB keyboards, and USB mice directly to a server |
| 7 |
>> with multiple video cards. I found a motherboard with 6 PCI-E slots: |
| 8 |
>> |
| 9 |
>> http://www.newegg.com/Product/Product.aspx?Item=N82E16813128508 |
| 10 |
>> |
| 11 |
>> 6 video cards could be installed for 6 workstations if the server goes |
| 12 |
>> headless, and even more if multi-headed video cards are used. Xorg |
| 13 |
>> requires some special configuration for this but this discussion from |
| 14 |
>> 2010 sounds like it's something that is actually done: |
| 15 |
>> |
| 16 |
>> http://forums.gentoo.org/viewtopic-t-836950-start-0.html |
| 17 |
>> |
| 18 |
>> These guys got it working in 2006: |
| 19 |
>> |
| 20 |
>> http://www.linuxgazette.net/124/smith.html |
| 21 |
>> |
| 22 |
>> 2. Set up a separate thin client for each workstation and run LTSP on |
| 23 |
>> the server. This seems inferior to #1 because it requires setting up |
| 24 |
>> and maintaining the LTSP server and client configuration, NFS, xinetd, |
| 25 |
>> tftp, dnsmasq, and PXE-boot. Bandwidth would also be limited compared |
| 26 |
>> to #1 and hardware and power requirements would be much greater. |
| 27 |
>> |
| 28 |
>> 3. Run a Plugable thin client for each workstation: |
| 29 |
>> |
| 30 |
>> http://www.amazon.com/dp/B004PXPPNA |
| 31 |
>> |
| 32 |
>> This likely requires running "Userful Multiseat Linux" on my server |
| 33 |
>> which is only packaged up for Ubuntu. The Plugable thin client |
| 34 |
>> connects to the server via USB 2.0 which makes me wonder if it could |
| 35 |
>> be made to work without Userful Multiseat Linux as a USB video card |
| 36 |
>> and input devices, but I imagine drivers for the video card and |
| 37 |
>> bandwidth over USB could be a problem. |
| 38 |
>> |
| 39 |
>> I think #1 is the way to go but I'd love to hear anyone else's opinion |
| 40 |
>> on that. Has anyone here ever set up multiseat in Xorg? |
| 41 |
> |
| 42 |
> Can you rely on Xorg devs to ensure that they are not going to break your |
| 43 |
> multiseat system in the future? |
| 44 |
|
| 45 |
Maybe I'm wrong, but I don't know why there would be (much) more |
| 46 |
likelihood of regression with Xorg multiseat than with anything else, |
| 47 |
including LTSP and all of its dependencies. In the context of both |
| 48 |
hardware and software, I think there are much fewer points of |
| 49 |
potential failure with multiseat than with an LTSP thin-client |
| 50 |
arrangement. |
| 51 |
|
| 52 |
> Are you sure that you will come across bandwidth issues if you follow option |
| 53 |
> #2? On a gigabit network at work we're running thousands of thin clients |
| 54 |
> distributed across hundreds of VM servers, and there is no noticeable latency |
| 55 |
> (unless a particular VM MSWindows server plays up). |
| 56 |
|
| 57 |
I'm sure I wouldn't. I only mentioned the increased bandwidth of |
| 58 |
multiseat vs. thin-clients as a technicality. |
| 59 |
|
| 60 |
> I understand that managing multiple boxen is always a greater burden, but |
| 61 |
> something like GNAP may lighten the work needed? |
| 62 |
> |
| 63 |
> http://www.gentoo.org/proj/en/base/embedded/gnap-userguide.xml |
| 64 |
|
| 65 |
That looks cool, but from my perspective it's another layer to learn, |
| 66 |
install, configure, and manage. chef and puppet take a different |
| 67 |
approach to lessening the burden of administrating multiple systems, |
| 68 |
but in the end neither approach comes anywhere near the hardware and |
| 69 |
software simplicity (and corresponding ease of setup and maintenance) |
| 70 |
of multiseat. |
| 71 |
|
| 72 |
- Grant |
Archives