Path: utzoo!utgpu!jarvis.csri.toronto.edu!mailrus!wasatch!cs.utexas.edu!uunet!ncrlnk!ncr-sd!hp-sdd!hplabs!hpfcdc!hpfclp!diamant
From: diamant@hpfclp.SDE.HP.COM (John Diamant)
Newsgroups: comp.windows.x
Subject: Re: Why X Windows?
Message-ID: <9740090@hpfclp.SDE.HP.COM>
Date: 25 Apr 89 05:40:30 GMT
References: <17319@cup.portal.com>
Organization: HP SESD, Fort Collins, CO
Lines: 36

>     1) on a single machine what are the advantages to multiple applications
>     sending messages to a window manager?
> 
> In some ways it's the other way 'round: there are no significant
> disadvantages, and it permits a network split (that's the big advantage).
> Explicit advantages would have to be "compared with what?".  It depends on
> a fair number of factors.

I can't agree with this statement.  There are significant advantages to the
client-server model, but there are disadvantages, the largest being
performance.  Typical X Toolkit applications, for instance, generate
incredible amounts of server traffic, and can be profiled as spending most
of their time in I/O on the X socket.  Now, it's true, much of this can
be blamed on the Toolkit and not the X protocol, but not entirely.  For
instance, popping up transient windows is markedly slower under a
window manager that uses SubstructureRedirect than one that doesn't.
The reason for this is INHERENT in the X protocol: multiple server round-trips
are required for a reparenting window manager.  In contrast, a non
client-server windowing system would be able to provide this negotiation
much faster.

I know your first objection to my statement would be that you can always
use fast communication on the local machine via shared memory or such, but
that really only provides significant benefits for large data transfers.
The process switching and synchronization delays would still exist.
Basically, X is a thin-wire protocol and that costs something.

Don't get me wrong -- I think X is great and the client-server model makes
a lot of sense for the benefits; I just can't agree with the statement
that there aren't significant disadvantages too.


John Diamant
Software Engineering Systems Division
Hewlett-Packard Co.		ARPA Internet: diamant@hpfclp.sde.hp.com
Fort Collins, CO		UUCP:  {hplabs,hpfcla}!hpfclp!diamant