[Author Prev][Author Next][Thread Prev][Thread Next][Author Index][Thread Index]

Re: [OT] Re: gEDA-user: PCB GTK version...

Samuel A. Falvo II wrote: 
On 7/26/05, Jakub Piotr CÅapa <jpc@xxxxxxxxxxxxx> wrote:

What? IP-domain? And what about SHM and Unix sockets? The network
transparency of X Windows is not the source of poor performance...


Actually, considering how verbose X11 protocol is "over the wire," yes
it is.  However, you are correct that X11 itself isn't the problem
here.  GTK is more than an order of magnitude slower than raw X11 (as
aptly demonstrated by comparing PCB and gPCB's performance on the same
hardware) is.

X11 is verbose but function calls aren't free either (especially when there are context switches involved) so I don't think the usage of sockets (UNIX or TCP/IP) is a real problem. The inefficient usage of them although probably is. I have seen comments on the inefficiency of modern toolkits and the amount of round-trips they require (when they aren't really needed) rendering the whole X11-as-a-networking-protocol almost unusable and killing the really cool thing the X11 architecture is.

Btw. The pure Xlibs solutions aren't ideal either. Killing the app simply because the connection to the X server broke kind of sucks. (and is really absurd because one of the key points in the wish list when the protocol was designed was to actually allow you to move apps between displays without an effort).

Also, the presence and utilization of SHM are negotiated via the
socket.  I believe, IIRC (and I could be wrong), that ONLY data is
transferred via SHM.  Actual X protocol commands are still sent via
the socket.

AFAIK you are right but that still makes quite a big difference (sending images over the pipe almost certainly would result in big lags).

--
Regards,
Jakub Piotr CÅapa