[c-nsp] PPPoE termination on 2620

[Dennis Peng]

Robert E.Seastrom [rs at seastrom.com] wrote:
> 
> Dennis Peng <dpeng at cisco.com> writes:
> 
> >> Does a PPPoE(oE) interface eat one IDB, or two?  I'm doing some
> >> similar stuff (running PPPoE(oE) on
> >> non-supported-but-the-code-is-there platforms) and watching my CPU, so
> >> I'm kinda curious...  "show idb" shows that it eats a hardware IDB,
> >> but is there a software IDB in play too?
> >
> > It uses one hwidb and one swidb. The IDB limit applies to the swidb,
> > and since the number of swidb is always equal or greater then hwidb's,
> > swidb's are the ones you need to worry about.
> 
> So, I guess I'm unclear on the concept of IDBs - do hardware IDBs
> exist in the shadow of the software IDB that is associated with the
> interface, or am I getting that wrong?  Has anyone done a Powerpoint
> for Networkers that shows how that part of the equation goes together?

The hardware IDB, is tied to the, well, hardware, the physical port. A
software IDB is created for each instatiation of an interface tied to
that physical port. There's always a main interface, but there could
be sub-interfaces too. Each sub-interface would use up an additional
software IDB.

> Speaking of things that I may not be entirely right on, I've been
> operating under the supposition that the majority of the sizing
> equation for PPPoE terminations was the amount of bandwidth (both bps
> and pps) one was pushing being the biggest deciding factor, with the
> number of sessions one is terminating playing a minor (but possibly
> non-linearly more significant as the number grows) role in terms of
> resource exhaustion.  I've also assumed that this assumption is
> correct not only in straight up PPPoE terminations but in LAC and LNS
> L2TP terminations as well.  Am I off in the weeds here, or am I
> substantially correct?

For aggregation of broadband sessions, the CPU is usually the limiting
resource, not the IDB limit because of the traffic patterns associated
with broadband subscribers. If you are aggregating narrowband sessions
(modems), then the traffic load is a lot less and in that case the IDB
limit may come into play before the CPU is overloaded.

Dennis

> Any recommended reading, lessons-learned white papers, or anything of
> that ilk would be appreciated.
> 
> Thanks,
> 
>                                         ---Rob
>