Re: AW: SV: focus failure with x11 and Mac OS X 10.5
Re: AW: SV: focus failure with x11 and Mac OS X 10.5
- Subject: Re: AW: SV: focus failure with x11 and Mac OS X 10.5
- From: "Andrew J. Hesford" <email@hidden>
- Date: Thu, 29 Nov 2007 10:10:23 -0600
On Nov 29, 2007, at 8:51 AM, Ambrose Li wrote:
On 29/11/2007, Andrew J. Hesford <email@hidden> wrote:
2. The CPU time for SSH tunneling is just offloaded into extra
burden on the remote system, which has the responsibility
of rendering an entire display for users. Since these
remote systems tend to serve multiple users, this issue is
compounded.
This is untrue. However you mean by "rendering the entire
display" (whether sending the X commands or acting on the X
commands), the situation is identical as the situation where you
use an ssh tunnel. The remote system still has to send all the
X commands, and the local system still has to act on all the X
commands to do rendering.
The cpu cycles and memory used for ssh tunnelling are extra,
but, as you correctly point out, you also reduce the network
bandwidth. But as to loading on the remote system, if you have
tried running programs over ssh vs over plain X, you need to
agree that the loading is less without ssh.
I guess we got sidetracked with SSH, because SSH tunneling offers more
than just remote displaying. It offers encryption, compression and the
convenience of not having to explicitly set DISPLAY on the remote
server or worrying about firewalls and X11 access lists. If SSH
tunneling is undesirable, you can always do a direct X11 connection
without XDMCP. This is the point I was trying to make: using XDMCP to
make the server draw and manage a whole display is more expensive than
having the server run a few client programs that connect to a remote X
server responsible for drawing and management. But SSH tunneling goes
above and beyond basic remote displays.
What happens when you move windows around in your XDMCP display? The
server has to redraw things, because it is responsible for keeping
track of that stuff. Without XDMCP, a proper windowing system managing
windows locally won't need to tell the remote system that something
needs refreshing every time you move a window. The remote system,
without XDMCP, doesn't care whether you move the windows, or even
where they are located.
In fact, because OS X leaves the final rendering to the compositor,
the X11 implementation should, in theory, never have to think windows
are obscured, and should never have to refresh windows as you move or
raise them. This is because X11 can treat every window as fully
visible, and leave the compositor to deal with which parts are really
shown. In this case, the remote windows should only need redrawing
when the user or client program makes a visible change to the window,
like inputting or displaying new information.
I say "in theory," but I don't know if the X11 server actually leaves
the obscuring to the compositor.
This is fair. I happen to disagree with this in general, but
within the context I agree with you. In any case, xdmcp not
working properly is a bug that needs to be ultimately fixed.
Since I don't know why an application would "require xdmcp" (I
cannot think of any reason), what I say is useless. But if an
application works with xdmcp and doesn't work without xdmcp,
then it looks like something is done in the user's xsession
which should be moved over to a startup script for the app. On
another hand, if the app works without ssh but doesn't work with
ssh, maybe it's the new trusted vs untrusted forwarding thing...
If it's broken, it should be fixed. I agree with that. Somebody
mentioned Oracle as something that doesn't work properly without
XDMCP. I don't know if this is correct, because I avoid XDMCP.
Regardless, my position is just that XDMCP ought to eventually be
allowed to die. I think it would be a good thing for users to become
accustomed to manipulating windows without relying on XDMCP to provide
a full "desktop" for each remote system.
--
Andrew J. Hesford <email@hidden>
Department of Electrical and Computer Engineering
University of Illinois at Urbana-Champaign
_______________________________________________
Do not post admin requests to the list. They will be ignored.
X11-users mailing list (email@hidden)
This email sent to email@hidden