[c-nsp] PPPoe Access Concetrators

[Robert E.Seastrom]

Alexandre Snarskii <snar at paranoia.ru> writes:

> Does anybody has real-life experience with 2800 in this role ? 
> I'm planning to test 2800 as an pppoe concentrator and wish to 
> know how much traffic/sessions it will be able to handle.. 

I've been thinking the same thing, but the 2811 only has 1400 software
IDBs.  Could get a little tight for us, I think.  Still, the price is
right...  would be nice as a bootstrapping platform until there's
enough money to support getting a few 7301s.  Do available feature
sets for the 2800 actually support acting as a PPPoE/L2TP server (LNS)?

I've used both 7206VXR/NPE300 and 7301s for this application, btw.
*with our customer traffic loads* (some of our customers are
filesharin-fools, but I hear that's a lot more common these days),
here's a sample of what we're seeing on two of our boxes right now:

7206VXR/NPE300 - 1396 connected users, 58% cpu, 58 mbps/13600 pps in+out
7301 - 1302 connected users, 22% cpu, 93 mbps/20200 pps in+out

(yes, they serve different markets, and the demographic difference shows!)

I'm figuring that running these any hotter than 75% cpu is probably a
bad call, but there are at least some sample real-world numbers to go
by there.

Soapbox:

The world could certainly do with an LNS/LAC/tunnelswitch platform
that ran on commodity PC hardware.  Plenty of CPU and RAM bandwidth to
do a few hundred megs of bit-banging, stripping headers off and
putting them back on again.

Unfortunately, just as is the case every time someone starts talking
about using PCs as core routers for small networks, the devil is in
the details and feature sets that one needs to get the job done.

For instance, Mikrotik RouterOS claims support for both L2TP and
PPPoE, but I had trouble getting PPPoEoL2TP working properly; it seems
to think of L2TP mainly as a vpn protocol, not as a generic
unencrypted transport.  No big deal that it didn't work, though, since
I'd already identified a deal-killer - RouterOS support RADIUS
Attribute 11 (Filter-ID), but not Attribute 242 - Ascend-Data-Filter.
Since we need to load filters on an ad-hoc basis, the "name the number
of the preexisting filter spec" approach doesn't work.

                                        ---Rob