jsparkes@bcars85.bnr.ca (Jeff Sparkes) (05/29/90)
I've been doing some measurements of packets from my NCD X terminal. Various window managers seem to generate enormous amounts of traffic for common operations. Popping up a menu or placing a window causes the NCD to generate packtes as fast as it can, about 300 packets/s. I realize that this comes from X's workstation heritage, where WM traffic was just memory copies. Does anyone know how to (or plan to) reduce this traffic? On our large network, it's possible for WM traffic to take up a large chunk of network bandwidth. DEC may have done the right with their new X terminal by putting the WM in the server. Do any other manufacturers (especially NCD) plan to do this? It would probably be better to have multiple loadable WM's, but I'm sure we could live with a single hardcoded one. -- Jeff Sparkes jsparkes@bnr.ca Cat: Is that what I think it is? Lister: What do you think it is? Cat: A big orange whirly thing in space!
rws@EXPO.LCS.MIT.EDU (Bob Scheifler) (05/30/90)
Does anyone know how to (or plan to) reduce this traffic? twm has been fixed, since R4, to no longer use a constant QueryPointer loop for mouse tracking. [Don't ask about public patches.]
toml@ninja.Solbourne.COM (Tom LaStrange) (05/30/90)
> I've been doing some measurements of packets from my NCD X terminal. > Various window managers seem to generate enormous amounts of traffic for > common operations. Popping up a menu or placing a window causes the NCD to > generate packtes as fast as it can, about 300 packets/s. I realize that > this comes from X's workstation heritage, where WM traffic was just memory > copies. > Does anyone know how to (or plan to) reduce this traffic? On our > large network, it's possible for WM traffic to take up a large chunk of > network bandwidth. > DEC may have done the right with their new X terminal by putting the > WM in the server. Do any other manufacturers (especially NCD) plan to do > this? It would probably be better to have multiple loadable WM's, but I'm > sure we could live with a single hardcoded one. Much of the network traffic is due to historical reasons. Older workstations such as the Vaxstation I and II were not fast enough to supply the WM with MotionNotify events sufficient to keep move/resize outlines tracking the mouse. Because of this, some window managers will send off QueryPointer requests rather than letting the server supply MotionNotify events. This makes performance look much better on slower workstations. Several people have mentioned that this is a "bug" and should be fixed. My how times change, the first versions of twm used MotionNotify events until I got a "bug" fix that implemented the QueryPointer stuff. Since no one out there is using a slow workstation, prehaps we should change it back :-) BTW, NCR also has the capability to put clients such as the WM into the terminal. -- Tom LaStrange Solbourne Computer Inc. ARPA: toml@Solbourne.COM 1900 Pike Rd. UUCP: ...!{boulder,sun}!stan!toml Longmont, CO 80501
jim@EXPO.LCS.MIT.EDU (Jim Fulton) (05/30/90)
Does anyone know how to (or plan to) reduce this traffic?
twm has been fixed, since R4, to no longer use a constant QueryPointer
loop for mouse tracking. [Don't ask about public patches.]
For those who really care, the fixes came from the patches John Carey
(john@labtam.oz.au) posted several weeks ago.mouse@LARRY.MCRCIM.MCGILL.EDU (der Mouse) (06/04/90)
>> DEC may have done the right with their new X terminal by putting the >> WM in the server. [...] It would probably be better to have >> multiple loadable WM's, but I'm sure we could live with a single >> hardcoded one. For what it's worth, I couldn't. > Much of the network traffic is due to historical reasons. Older > workstations such as the Vaxstation I and II were not fast enough to > supply the WM with MotionNotify events sufficient to keep move/resize > outlines tracking the mouse. Because of this, some window managers > will send off QueryPointer requests rather than letting the server > supply MotionNotify events. This makes performance look much better > on slower workstations. Isn't this what pointer motion hints are all about? My window manager used to depend on MotionNotify, but performance over a slow network was very bad (the tracking would fall behind the mouse, though it would catch up if you let the mouse sit still). I changed it to use motion hints and it works fine now. der Mouse old: mcgill-vision!mouse new: mouse@larry.mcrcim.mcgill.edu