[pptp-server] pptpd still not handling dropped connectionsproperly?

[tmk]

> Matt, Kevin, is pptpd ok for > 50 users. I noticed some potential leak
> fixes in the last changelog. Would this many users eat up the system
> resources? How well does it scale? Anybody know?

pptpd should scale really well. It simply creates a new pppd, gre, and
pptpctrl process for each call, so as long as your system can handle that
load, everything should be peachy. I haven't done any sort of memory usage
anaylysis on any of our programs, but they seem to be fairly small. The only
thing that would slow us down is that most older (2.0.x?) kernels only
support 64 network devices (ie ppp?, eth0, etc) so keep that in mind.

i have been planning on optimizing the ctrlpacket.c section of code to use
(much) less processor power, at the expense of more memory. Is that  a
reasonable trade off or do we want more free memory and a heavier load on
the cpu? i'm open to comments.

Also, the latest versions of pptpd should have some built-in link status
detection and such, so if the link goes down, it should AT LEAST kill off
gre. So to Bitt, download the latest ver of pptpd and see if that solves
your problem.

Last thing is we don't have a really clean way to kill a pppd connection.
anyone know of a way to do it? the problem is that pptpd daemonizes itself
and gets a new pid, leaving us with no way to kill it. we could get the pid
from the lockfile, or we might be able to use nodetach, but there must be a
better way.

thanks for helping out. I'm surprised to hear that our product is working
better than NT. I know the win9x pptp client has real problems, but i
figured NT would be better. Go opensource :)

Kevin