The MPLS WG Archive

["Cheng-Yin Lee" <Cheng-Yin.Lee@alcatel.com>]

Kireeti et al,
It looks like separating the issues may be useful, for e.g. something along the
line:

For "ping" mode:
1) send a labeled test pkt on a p2p LSP
2) send a labeled test pkt on a p2p LSP, with PHP (no label after PHP)
3) send a labeled test pkt  on a mp2p LSP (with and without ECMP)
4) send a labeled test pkt on a mp2p LSP, with PHP (with and without ECMP)
5) FRR case, etc...

Similarly for "traceroute" mode.

This may not be the best/correct/complete way to separate the issues, if so pls
suggest other ways. But I think separating the issues may help in progressing the
draft in the WG. One may find that it may not be necessary to use 127/8, or that
the problem could be partitioned.  Just a thought ...

thanks,
cheng-yin

Curtis Villamizar wrote:

> In message <20021211023339.K19625-100000@kummer.juniper.net>, Kireeti Kompella
> writes:
> > Hi Curtis,
> >
> > On Fri, 6 Dec 2002, Curtis Villamizar wrote:
> >
> > > Maybe one of the authors can join (do they really want to?)
> >
> > Not really :-)
>
> Smart move.
>
> > > whether they would see fit to add a sentence or deem it too obvious.
> >
> > Saying what?  "Setting the destination IP address to something from 127/8
> > causes some to break out in a rash, but is perfectly valid"?
>
> I'll take that as closer to a "no its too obvious".  My first choice
> as well.
>
> If there remains insistence on the list that we justify the choice of
> 127/8, the criteria would be:
>
>    An ideal destination address would be deliverable to the local LSR
>    when TTL expires, regardless of where in the topology that router
>    was.  This first requirement is so that an address on the next
>    MPLS-ping capable downstream LSR need not be known to get a
>    response.  An ideal destination address would be deliverable within
>    an MPLS LSP but otherwise unrouteable even in networks using the
>    IANA designated private address space [RFC1700].  This second
>    requirement insures that the MPLS-ping packet is not forwarded by
>    IP and does not travel further than the LSR where TTL expires, or
>    the egress LSR.
>
>    The address space 127/8 is explicitly chosen by IANA to be used
>    only to reach the local loopback device on a host or router
>    [RFC1812].  This address space is never routeable.  For these
>    reasons, destination addresses are chosen from the address range
>    127/8.  Multipath capabilities on some routers, are based on a hash
>    or other algorithm applied to IP source and IP destination
>    addresses [RFC2991].  In order to exercise multipath, addresses
>    from a range of addresses must be used.
>
>    An MPLS-ping capable LSR as an LSP egress must be able to deliver
>    packets exiting the LSP with an address anywhere in the range 127/8
>    and a UDP payload with destination port 3503 to a local loopback
>    device.  An MPLS-ping capable LSR as an LSP midpoint must be able
>    to deliver packets for which TTL expires which have an IP payload
>    and a destination address anywhere in the range 127/8 and a UDP
>    payload with destination port 3503 to a local loopback device.
>
> Maybe I should have suggested text in the first place.
>
> Now we have two questions...  First do you agree with the text.
> Second is do we need it.
>
> > Kireeti.
>
> Curtis