Comment 15 for bug 269082

David Bruce (davidstuartbruce) wrote :


I'm the maintainer for Tuxmath and Tuxtype. Below is what I sent out yesterday to our dev list:

"This is puzzling as the three programs don't really share much code
(the menu stuff from Tuxtype was adapted for Tuxmath, and the Comet
Zap activity in Tuxtype is an adapted version of Tuxmath, but that's
about it). I hate to just say "it must be SDL", but it seems weird
that three independently-coded programs would all create the same bug.

From reading the posts on the launchpad forum, it sounds like Tuxmath
is using all the cpu cycles on the server when run from a thin client.
 It certainly doesn't do this when run on a single machine - a quick
check with top on my desktop machine (amd64 dual 3800, 2 gig RAM)
shows that neither tuxmath itself nor Xorg exceeds about 3% cpu. Of
course it would use more on less powerful hardware, but I would expect
a typical server machine to be able to run quite a few instances of
tuxmath. I started up 10 instances just now on my desktop with no
discernable effect on system performance.

So, I don't have any immediate ideas. It might be helpful to know at
what point in the program the resources all get eaten up (menus vs.
game itself). Also, if the problem is 100% cpu utilization, it could
be very informative to profile tuxmath with valgrind to see where all
of the cycles are going. Tim is our resident valgrind expert and he
might know how to do this."

So, I think someone with a running network needs to get some more specific info about what is happening to bog the system down (cpu utilization vs. network saturation, where in the programs the problem is seen, etc) and ideally do some actual profiling. The programs do not tax the machine much at all when run locally, even on old, slow computers, so I suspect that something is happening in either SDL or Xorg functions related to the remote X session that is slowing the server down.

David Bruce