The context switches of pppd aren't that bad. The memory consumption is a problem, but the context switches aren't a problem.
Well... 2000 context-switches per second with one call active, just transfering the Linux kernel source via scp. Though I guess that's partly to do with the scp bit. I really need to get the test machines working as routers (like they're supposed to) and not as end-points.
> Yeah, I wasn't really keen on some of the implementation description on > that one...sounds like it might work pretty well...but the description > kinda bothers me.
The code is scary to look at - he's implemented only as much of PPP and L2TP as he's needed, in order to get his ISP working with one particular type of LAC (portmaster, IIRC). But it works for him.
Just a an example, the code only implements LCP echoes. It doesn't look like it can negotiate most common options in IPCP (I didn't see anything about MTU, for example). But I've only just glanced through the code very briefly.
I'm hoping to have time to look at it over the weekend (hah!) and see how the three implementations get on together.
Alex