Subject: RE: Userland PPP
To: Jason Thorpe <email@example.com>
From: Martin Husemann <firstname.lastname@example.org>
Date: 02/19/1999 09:55:00
Jason R. Thorpe wrote:
> In NetBSD, the PPP connection setup protocol _IS_ handled in userland
> by pppd, but data packets (i.e. ones that don't have the control bit
> set in PPP's "ethertype" equivalent) are packaged into mbufs and handed
> directly to the appropriate next layer.
> NetBSD's method is MUCH more efficient, which will be important as we
> implement PPP over faster media, such as Ethernet (yes, really; there
> is an internet draft for this) and ATM.
I wholehertedly agree!
And I can add a little story: in ISDN4BSD we have completly kernel base PPP
support (i.e.: no pppd involved). I considered this a bad hack and tried to
prevent it when it was first introduced.
I then installed a test version on our 4MB i386 ISDN router (I think
'powered' by an 486SL Cyrix chip at 33 MHz) and PPP link establish time
improved from 15s-20s to 1s-4s (from first packet causing a dial request to
first packet actually answered by remote).
I suppose this machine would not have been able to drive two simultaneous
connections with a userland implementation at full speed. And of course
there was swapping involved when firing up pppd...