PDA

View Full Version : True multi user X Windows Linux


yitzhakbg
02-26-03, 11:32 AM
What is the status of work on implementing multiple X Window video consoles for simultaneous users on Linux running on a single CPU? We want to implement a classroom environment for multiple players on graphics heavy games, but on a low budget. Any pointers for implementation? The Linux Console Project on SourceForge is defunct. I was the following implementations:
http://cambuca.ldhs.cetuc.puc-rio.br/multiuser
http://startx.times.lv/
We would like to use multiport video boards to save motherboard slots. Which Nvidia cards would be best and cost effective?

sanat
02-26-03, 12:18 PM
The response of multiple consoles gets too choppy under your shceme. It is much better to make Xterminals out of Nforce boards, and then run the game on an application server. This method eliminates the choppy response.

try http://ltsp.org

and

http://pxes.sourceforge.net/

yitzhakbg
02-26-03, 01:07 PM
Wouldn't the choppiness response of multiple consoles depend on the machine's processing power and quantity of memory? We tend to think in terms of using older, less powerful computers as X terminals with a powerful application server. Don't you think it would be fair to assume that a relatively powerful Linux box with multiple consoles could provide more bang for the buck than the application server and its' satellite X terminals?

sanat
02-26-03, 04:41 PM
No. No matter how fast a CPU is, there can be demands that greatly impact usage. When the CPU usage goes up, user interface responsiveness changes. You can see for yourself by compiling the Linux kernal on an xterm while surfing the web or using gimp or some other activity that requires mouse and keyboard.

CPU usage is full of peaks and valleys, even though most of the time the CPU is loafing. When the CPU is at a short-lived peak everybody's mouse and keyboard will be effected.
http://swnews.net/folder/snapshot2.gif

Using Xterminals, the mouse and keyboard consistency aren't subject to the application server load. I realize this would be more expensive, but not all that high. An Nforce at US$78, a Duron at $32, 128mb memory at $15 and then a cheap case. No disk drives, CDROMs, or floppys needed.

r0gu3
03-09-03, 10:21 PM
Well actually you demonstrate exactly why it is ok to have multiple users on the same computer... All you have to do is simply overlay some usage graphs (in random starting places) several times, and you will realize that it is very rare that multiple users(my example 3) will have a peak at the same time (less than 99.5% of the time, although it is cpu dependant)... so basically the users will only see latency about .05% of the time, and only for an extremely short period of time... with some process priority controls its really easy to administrate also...

I am working on a paper (and a presentation for the local lug) on how to setup (and manage) a system for multiple local users... the links which you gave only work with certain hardware (they can get unstable ESPECIALLY with multiple geforce1/2/3 cards in the system) but simply using different Xservers, you can get around all their limitations.

- r0gu3

sanat
03-10-03, 12:41 PM
When you're running multiple users with multiple cards, keyboards and mice all plugged in the same machine, ALL users feel ALL jumps in CPU usage.

When you're running the same number of users via Xterminals, none of the users feel any of the jumps in CPU usage.

r0gu3
03-17-03, 12:01 AM
Well I don't want to insult you... but the part i think you don't understand is that you can only "Feel" a cpu spike if you need to use those cycles...

Try overlaying your graph on top of itself in random starting points... basically your users can "Feel" the other users is when the summed graphs are above 100%... So basically YES it is possible for the to feel the other users, but the probability of it happening is incredibly small (so small most users won't notice it, and there in can be neglected)...

One of the underlying pricipals this exploits is the fact that most desktop pc's spend most of their cycles WAITING for user input...

In practice... a thin terminal exploits some of the same principals to save network bandwidth (depends on implentation)

I think that my real point here isn't that people shouldn't use thin clients, but that having multiple users on a single pc can save people A LOT of money in both hardware AND administrative costs compared to thin clients. I don't think this technique can be used to replace most thin clients, but it gives REAL options to implementors.

- r0gu3