Re: suggestion.

[Danial Thom <danial_thom@xxxxxxxxx>]

--- Paul <mirnshi@xxxxxxx> wrote:
> Joerg Sonnenberger <joerg@xxxxxxxxxxxxxxxxx>
> wrote in
> news:20050503183906.GD94744@xxxxxxxxxxxxxxxxx: 
> 
> > On Tue, May 03, 2005 at 03:55:35PM +0000,
> Paul wrote:
> >> 1. Like FreeBSD, the ppp is still in
> DragonFly. But the version is too
> >> old. Why not choose mpd? More fast and
> flexible, and based on
> >> netgraph. 
> > 
> > netgraph is not neceseary is a good thing :)
> > 
> PPP eat more CPU than mpd. In heavy gateway,
> such as pptp/pppoe, mpd is 
> better choice.
> Unlike linux, PPP of BSDs is in the user space.

Uh, well you seem to have a twisted view of the
subject; nothing personal.

1) Netgraph IS bad. Its counter-intuitive and
slow. Sort of like streams that never caught on
with any other OSes. Its existence caused FreeBSD
to never do anything the "right" way, because the
company that supports netgraph always did it
first.
2) MPPP (mpd is a daemon, not a protocol) is MUCH
less efficient than ppp. In its pure form, ppp
has no overhead; after initialization there is no
required overhead. The "cpu usage" you see is
likely in the implementation of echos, which are
completely optional and should be disabled. Most
unix implementations, because written by "some
dude in the ural mountains", do the echos in user
space, which is just plain stupid (however easier
than doing them in the kernel, the way it should
be done). 
3) MPPP has an archaic requirement that packets
arrive in order. This requires front end queues
similar to fragment processing, that is a waste
of memory, cpu cycles and results in slowed
connections, since all modern TCP stacks can
handle out of sequence packets.
3) MPPP was designed for slow, dial-up lines (and
ISDN). Per packet load balancing gives better
performance in the real world, with lower CPU
usage.

Danial


		
Yahoo! Mail
Stay connected, organized, and protected. Take the tour:
http://tour.mail.yahoo.com/mailtour.html