On Thu, Jul 06, 2006 at 12:52:23PM +0200, [EMAIL PROTECTED] wrote: > I wont blame the developers because they offen said the userland pppoe has > a lot overhead but that`s simply a bad joke. :-(
Userland pppoe copies the pppoe session frame from kernel to userland via bpf(4) into the pppoe(8) program which then writes it back to the kernel (into a pipe) and back to the ppp(1) program and then back into the kernel via the tun(4) interface, it will then be recv(2)'ed from the proper program that has a session/connection. So you're bouncing around from kernel to userland back to kernel back to userland back to kernel back to userland for receiving just one frame of data. *ponder* I'm not sure if a context switch can be saved in this execution path if between pppoe(8) and ppp(1) a shared memory segment exists, and packets are copied that way. It could speed things up perhaps. With kernel pppoe(4) the pppoe session frame enters the kernel and gets recv(2)'ed from userland. A lot less switching back and forth from userland to kernel. > But I hope the kernel-pppoe will provide me nearly the same bandwith like > the Windows-pppoe-Tool. :-) > So if you4ve a DSL-connection wich is maybe faster then 4Mbit.. use the > kernel pppoe (even I don4t know how good it performs). :) Good luck, -peter -- Here my ticker tape .signature #### My name is Peter Philipp #### lynx -dump "http://en.wikipedia.org/w/index.php?title=Pufferfish&oldid=20768394"; | sed -n 131,136p #### So long and thanks for all the fish!!!
