Discussion Thread: Thin Client X Window Systems

[Jim McQuillan]

On Tue, 5 Apr 2005, Daniel Stone wrote:

> On Mon, Apr 04, 2005 at 10:05:51AM -0700, Stuart Kreitman wrote:
> > This message is intended to start a thread of discussion of X issues
> > that pertain to reduced or
> > non-Linux and non-Unix desktop environments, ie LTSP, X-on-Windows,
> > SunRay, VNC, X terminals, embedded X.
> >
> > These environments are in danger of being ignored or outright crippled
> > by the onslaught of major feature
> > advances that we see happening on the mainstream Linux and Unix
> > desktops, yet, there are
> > considerable populations of users and developers for whom
> > one-organism-one-OS is not a preferance.
> >
> > We need to develop ideas that advance these populations as well.  Issues
> > are easy to find: better management
> > of heavy high and low bandwidth networks, the effect of many Gnome and
> > KDE class desktops on shared
> > central servers,  hot desktopping,  remote device attachment.
>
> The question being, should we attempt to manage the network side of it?
>
> LBX hasn't provided many tangible advantages in terms of being able to
> usefully run a modern desktop over low-bandwidth (and/or high-latency)
> links.  The two projects which have shown the most promise are external:
> SSH and NX.

I believe keithp is the author of LBX and even he will tell you that LBX
is crap.

SSH works nicely, until you try to run lots of thin clients off a single
server.  I've got a server with 140 thin clients.  Pushing all that
traffic through ssh, which is out-of-process, would kill that server.

>
> NX seems the most promising, although there is funny licensing abound in
> that project, and we can't just merge it straight in.

I'm hearing very good things about NX, but like you say, the licensing
could be a problem.

>
> So, my question is, should we try to encourage and develop
> low-bandwidth/high-latency solutions, or should we just sit back and let
> SSH and NX keep on doing this?
>
> :) d
>

Are we convinced that SSH and NX have the problem all sewn up?  Or are
there better solutions waiting to be discovered?  Maybe there's
something we can do in cooperation with those teams to get to the
next level.

Certainly the study that Jim Gettys and Keith Packard did on startup
times of an X client is helpful in understanding the problem of return
trips in the X protocol.

I think that's the purpose of this working group.  To bring the issues
to light, so we can toss around ideas, and either identify existing
solutions, or form a plan to build solutions.

Jim McQuillan
jam at Ltsp.org