[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Xouvert-general] Network transparentcy and modules

From: James Hoegl
Subject: Re: [Xouvert-general] Network transparentcy and modules
Date: Tue, 19 Aug 2003 09:13:30 -0500
User-agent: Mozilla/5.0 (X11; U; Linux i686; en-US; rv:1.2.1) Gecko/20030225

James Best wrote:

Jonathan Walther wrote:

On Tue, Aug 19, 2003 at 01:05:45AM +0000, Mark C. Ballew wrote:

We *cannot* remove network transparency. Period. There is absolutely no question about it whatsoever. X11 *is* a _network_ protocol. I think the reason everyone thinks its slow, is because everyone thinks that X11 on a local machine goes thru tcp/ip. It doesnt. It uses UNIX sockets which are very fast.

Oh no, this totally isn't what I'm getting at. I was suggesting that the
network transparency be even more transparent, by allowing different
network protocols to be plugged in. I know to an extent this already
exists, but I don't know if it is hard coded in or easily to modify.

Next time check the code first.  Adding in new transports is relatively
easy; the actual transport specific code is small.  As someone else on
the list said, the code to support X over DECNET is only 6k.


Adding transports looks to be easy - but what about the protocol? I believe he said about "network protocols". I, too, am curious about this. I understand that X is the protocol, and without it it wouldn't be X. I have looked at the protocol, though, and it is far from the most efficient I have seen.

Although Unix pipes are quite fast, cutting the data sent over them in half could have a substantial impact on the speed of the system. As clients build layer upon layer of gui abstraction, the X transport is sending more and more little messages, and they add up. There my be better, faster encodings that can be developed.

On a different note, I don't think that X's sluggish response is from the protocol or the network. From my measurements and experience, it's from task switching. When I click on a button on the application, X gets the event, sends it to the client. The client does something with it, and usually responds with some kind of redraw of the widget to indicate the button was pressed. Then, the user releases the button, causing another event, another draw, and a number of other actions. This communication causes the computer to switch tasks many times for even the most basic button click. At this point, other than kernel tuning, I don't know how to reduce this. This is where I am most curious to help.


Xouvert-general mailing list

Everyone keeps stating how fast the Unix pipe is for the network transparency, but how fast is it? As it is now, the state of computer hardware is well beyond network protocols in speed and efficiency, for obvious reasons. So using a network protocol on a computer is not very logical in the course of speed and efficiency. If the Unix protocol can handle it, keep it, otherwise a new protocol, specifically for the computer is needed. Unfortunatly I am not as up on Unix specific protocol, what is it called exactly and what OSI layers does it consist of?
Ill do some checking while waiting for a response.
We will see which is faster ;)

James Hoegl

reply via email to

[Prev in Thread] Current Thread [Next in Thread]