On Wed, Apr 04, 2001 at 11:38:07AM -0700, Serge Maskalik wrote:
> Thus spake Jesper Skriver (jesper@skriver.dk):
>
> > On Wed, Apr 04, 2001 at 11:01:41AM -0700, Kireeti Kompella wrote:
> > > Hi Robert,
> > >
> > > > The only use of exp-null today would be to pass the QoS info in the EXP
> > > > bits to the tailend as otherwise this info is lost at PHP and tail has
> > > > no way of knowing how to schedule the packets (of course assuming that
> > > > QoS in IP header is different then the one in EXP field - something what
> > > > we call QoS transparency :).
> > >
> > > Ah, so you don't want to touch the IP QoS/ToS bits. Okay, got it.
> > >
> > > Somewhere in this conversation, there was mention that ciscos had
> > > trouble with label 0/label 3 (and ditto junipers). I can speak for
> > > the latter; can you comment on the former? If there is no issue with
> > > using label 0, should Juniper simply go back to using label 0 always?
> > > I'd rather not be putting any more knobs than are really necessary ...
> >
> > I think it's usefull to, on a per interface basis, be able to choose
> > between Explicit and Implicit null.
>
> Can you explain why it would be useful per interface?
We have a specific application where we on a number of dedicated routers
originate default route, for use of the routers in the network not
carrying full routing (mainly dial and DSL aggregation boxes), these routers
will use the LSP formed by LDP/TDP to get to these "default origination
boxes", PHP will occur, and the "default .. boxes" will recieve the
packets as ordinary IP packets, do NetFlow accounting on them, and do a
L3 lookup, and forward them on the LSP to the exit point ...
> I think it makes sense to do on per-LSP basis.
Much harder to control on a per-LSP basis, especially when talking about
LDP/TDP controlled LSP's.
> > So a knob would be nice to have.
> >
> > > What if the label stack has depth > 1? PHP in that case? Bit yucky,
> > > but doable.
> >
> > rfc 3032 section 2.1 state
> >
> > i. A value of 0 represents the "IPv4 Explicit NULL Label".
> > This label value is only legal at the bottom of the label
> > stack. It indicates that the label stack must be popped,
> > and the forwarding of the packet must then be based on the
> > IPv4 header.
> >
> > So yes, for label stacks with a depth > 1 you must use 3 and not 0
>
> I also realize the 3032 recommendation, but would
> you not want to maintain the EXP setting state
> regardless whether LSP hierarchy is involved or not?
It doesn't give any benefits, with multiple labels, you will still have
a lable to carry the EXP bits in, and when pushing multiple labels (or
pushing a additional label), the same EXP bits must be set in all the
labels, so doing PHP doesn't remove the possibility to queue based on
the EXP bits.
And then the ultimate router doesn't have to de 2 lookup's on the
packets.
But if the implementation is much easier by doing Explicit all the time,
then no alarm.
/Jesper
-- Jesper Skriver, jesper(at)skriver(dot)dk - CCIE #5456 Work: Network manager @ AS3292 (Tele Danmark DataNetworks) Private: FreeBSD committer @ AS2109 (A much smaller network ;-)One Unix to rule them all, One Resolver to find them, One IP to bring them all and in the zone to bind them.
This archive was generated by hypermail 2b29 : Mon Aug 05 2002 - 10:42:41 EDT