psmith@baynetworks.com
Low-Bandwidth X (LBX) attempts to recognize that in this day and age, not everyone will be a fast LAN hop or two away from the system that they are running their applications on.
The X protocol can generate an extraordinary amount of traffic, especially for simple-seeming things such as creating new windows. As anyone who has tried to use X over a dial-in modem at 28.8 or even higher can attest, creating new X windows can involve an excruciating wait.
LBX is fundamentally a compression and caching scheme designed to minimize the amount of X traffic generated between two systems.
As of the X Consortium's release of X11R6.3 in December, 1996, LBX is a full extension to the X protocol. For XFree86 folks, that's XFree86 version 3.3.
If you use a modem to dial into a service provider, then run X applications on remote machines with their DISPLAYs set to your local machine (or vice versa), LBX will speed up that connection. Also if you set DISPLAYs from systems across WANs (other countries, for example) or other slow links, LBX can help.
LBX is useless, of course, if you're only running applications locally, or if you're not running X at all.
Also, if you're running on a fast LAN, LBX won't be much help. Some people say "if LBX cuts down on network traffic, wouldn't it be good to use even on fast LANs?" It might be, if your goal is to reduce network traffic. But if your goal is to get better response time LBX probably isn't what you want. Although it does introduce caching and compression, that comes at a cost on both ends (extra memory for caching, and extra CPU for decompression). If your link is fairly speedy LBX will probably result in an overall slowdown.
LBX works by introducing a proxy server at the client side, which performs caching and compression. The X server knows that the client is using a proxy server, and decompresses accordingly.
Here's a normal setup for remote X clients. In our discussion, LOCAL is always the workstation sitting in front of you, whose monitor you're looking at, and REMOTE is the remote workstation, where the actual application is running.
REMOTE LOCAL
+-----+ +-----+
| APP |-\ Network +----------+ | |\
+-----+ \--------------------------->| X SERVER |=>| ||
+-----+ / (X Protocol) +----------+ +-----+\
| APP |-/ /_____//
+-----+
When using LBX, a proxy server (lbxproxy
) is introduced on the
remote side, and the applications talk to that process instead of
directly to the LOCAL server. That process then performs the caching
and compression of X requests and forwards them. It looks like this:
REMOTE LOCAL
+-----+
+-----+ +-------+ Network +----------+ | |\
| APP |->| PROXY |----------------------------->| X SERVER |=>| ||
+-----+ +-------+ (LBX/X Protocol) +----------+ +-----+\
+-----+ / /_____//
| APP |--/
+-----+
Details on exactly what caching and compression LBX does is beyond the scope of this document.
You need an X server on your LOCAL system which has the LBX extension compiled in. Unless you explicitly told it not to when building it, X11R6.3 servers automatically enable LBX. Also, all XFree86 3.3 servers have LBX enabled by default.
You can use the xdpyinfo
command to see if your server has the LBX
extension: run xdpyinfo
and look at the list just under "number of
extensions"; you should see "LBX" listed there.
Next, you need to get an lbxproxy
program compiled for the REMOTE
system. This is the tricky part. If the remote system is not the same
type as your local system, the lbxproxy
on your local system will
do you no good, of course.
There is unfortunately no "broken out" distribution of lbxproxy
, so
you will have to either (a) get and build most, if not all, of X11R6.3
for the remote system, or (b) find someplace to get a pre-compiled
lbxproxy
binary for your system. The latter is much simpler of
course.
The lbxproxy
is simply a single executable. There are no
configuration files, resource files, etc. associated with it.
The REMOTE system does not need a new X server (as always, the REMOTE system doesn't need any X server running).
The application you want to run does not need to be linked with any special version of X, or any special libraries; I regularly use commercial X11R5 apps over LBX with no trouble.
You do not need root or other privileged access on the REMOTE
system; the lbxproxy
process runs under your normal access
permissions. Further, you can run it right from your home directory: it
does not have to be installed anywhere.
OK, here it is... after all that it's actually quite simple. Replace LOCAL and REMOTE below with the hostnames of your local workstation and remote system, respectively (don't get them mixed up!)
On LOCAL:
xhost +REMOTE
. If you use
xauth
you may need to do more than this; see the xauth(1) man
page for more information.
You should consult the
Remote X Apps Mini-HOWTO if you're not familiar with remote X access permission
setup.On REMOTE:
lbxproxy
and tell it to forward to the LOCAL X server,
like this:
$ lbxproxy -display LOCAL:0 :1 &
This tells lbxproxy
to use display :1
on
the REMOTE system; if that system has >1 display already you can use
:2
or whatever instead.
lbxproxy
is providing, instead of the normal display:
$ DISPLAY=:1
$ export DISPLAY
Or, if you use csh or clones:
% setenv DISPLAY :1
xauth
you will need to ensure that your
cookie is available locally. See the
Remote X Apps Mini-HOWTO for more information on this.
That's it; all X apps that are started up pointing to :1
will use
LBX. Of course, there's no reason you couldn't also start X apps
pointing to LOCAL:0
and have both running at the same time.
Here are some common problems:
lbxproxy
exits with an "access denied" error.
This means the LOCAL system isn't accepting connections from the REMOTE system due to permissions errors. See the Remote X Apps Mini-HOWTO for details on these issues.
As a simple trouble-shooting measure, try running a simple X app
like xclock
on REMOTE and have it display on the local
system without using lbxproxy
:
$ xclock -display LOCAL:0
If that doesn't work, it's xhost
or some other basic X
problem, not LBX.
The only documentation available in a standard X distribution may be the lbxproxy(1) man page.
If you have access to the X source tree, then very interesting information on LBX is available there:
xc/doc/specs/Xext/lbx.mif
(Framemaker MIF)xc/doc/hardcopy/Xext/lbx.PS.Z
(Compressed Postscript)xc/doc/hardcopy/Xext/lbxTOC.html
(HTML)More detailed discussion of specific LBX algorithms is available here:
xc/doc/specs/Xext/lbxalg.mif
(Framemaker MIF)xc/doc/specs/Xext/lbxalg.PS.Z
(Compressed Postscript)If you don't have access to the X11 source, you can obtain these files from the X Consortium's FTP site.
If you don't like lbxproxy
for some reason: you're not
satisfied with the performance, it doesn't work for you, you don't want
to hassle with creating an lbxproxy for the remote host, or you simply
are interested in trying other options, there is at least one other
package for X protocol compression (anyone have others?)
dxpc
works in essentially the same way as LBX. However, to
avoid having to implement an X extension and modify the X server code,
dxpc
uses two proxies: one that runs on the REMOTE host, like
lbxproxy
, and one that runs on the LOCAL host.
The REMOTE host proxy communicates between the X clients and the LOCAL host proxy, and the LOCAL host proxy communicates between the X server and the REMOTE host proxy.
So, to both the X clients and the X server, it looks like X protocol as usual.
lbxproxy
.
The source for dxpc is available at ftp.x.org.
There is a WWW homepage for dxpc that gives a lot of good information, including pointers to the dxpc mailing list, access to the source code, and a number of pre-built binaries for various platforms:
http://ccwf.cc.utexas.edu/~zvonler/dxpc/
Ken Chase <lbxhowto@sizone.org>
notes that
ssh
can
be used for compression. Although its main purpose is to provide
security, it also compresses the data it sends.
Thus, if you run X over a ssh
link you will automatically obtain
some amount of compression.
I don't know. Both LBX and dxpc
are certainly better at raw
compression than ssh
. Of course, ssh
provides the added
advantage of security. And of course, there's no reason you can't use
both ssh
and one of the other two, to get good compression and
security.
It shouldn't be hard to run some benchmarking against these options and get both subjective and statistical measurings of performance. But I haven't done this, and I don't know of anyone who has.