just to throw in my $0.02:

i'm running rp-pppoe on a P90 with 8 mb ram (1.5ZA), and happily running
a dhcp server as well as named and sshd, and at one point I was also
running (though infrequently used) httpd, and I haven't noticed any real
problems.  This box is routing/splitting a verizon line for something like
6 computers, and the only problem i've had was a flaky rtk card (which
i've since replaced).  I'm doing an apt-get for linux kernel source on
one of my internal boxen and  have a sustainted transfer rate of around
90kB/s right now and rp is stablized out using around 35% of WCPU/CPU

of course, i had a real pleasant experience setting up in-kernel pppoe
as well, the only problem was the last time I checked, there wasn't
any way to get DNS server information from the pppoe provider.  so, unless
you happen to know that information (or are running your own nameserver, or
feel like being the one to add that feature), you probably want rp.
  

--sean

On Tue, Jun 25, 2002 at 04:58:38PM +0200, Martin Weber wrote:
> On Tue, Jun 25, 2002 at 11:20:34AM -0300, Jared D. McNeill wrote:
> >
> > If you're going to be using a slower machine, definitely go with the
> > in-kernel PPPoE. rp-pppoe on my AXPpci33 Alpha (166MHz) used all of my
> > available CPU (which became a bottleneck -- I wasn't getting full speeds
> > as a result). Switching to the in-kernel PPPoE, CPU usage dropped to ~3%
> > and I now utilize my full bandwidth.
> 
> This is very interesting. I use rp-pppoe on a pI 100, and the machine
> is mainly deep asleep. When I'm utilizing full bandwidth, CPU's at about
> 8-12 percent. Before the pI I used a 486 dx2/66, and even that one was
> not over at max 25 percent CPU. I wonder what was using that much CPU
> on your machine. I have of course ipnat and ipf running there, sshd
> and rp-pppoe. (this is a 768 kbit downstream dsl)
> 
> -Martin
>