The MPLS WG Archive

["Martin, Christian" <cmartin@gnilink.net>]

If I may comment here from an operational perspective...

I think the Hello mechanism, as specified, could be extended to occur upon
link initialization.  One of my most frustrating experiences with RSVP state
was when a new link was turned up with a fat-fingered authentication key.
Since there is no initial RSVP state acorss the link until a PATH message
needs to be sent, I was unaware of it until I had to troubleshoot a label
request across 13 routers.  Were it part of the specification to send
Hello's and build an adjaceny during link intialization, it would have been
something detectable during configuration.

Perhaps the draft can be extended to address this operational need?  How
difficult would it be to add:

"Upon link initialization, an interface MAY/SHOULD immediately send an RSVP
hello to the all routers multicast address on broadcast media, or to its
direct neighbor on ptp media.  For each RSVP hello received, record the
source IP address in the RSVP adjacency table.  Transmit hello's every
configurable interval."

Of course, this example discusses authenticated RSVP messages, but other
types of issues can impair RSVP without easy detection.  As it is today, you
can either send a test PATH message across the network to build state, or
you can hope it works when you need it.  Neither is operationally
acceptable, IMHO.

Regards,
 -chris


> >"Sanford, Bill" wrote:
> >> 
> >> David, wouldn't the lack of information from the absence if the
> >> RSVP LSP establishment almost guarantee that the RSVP Hellos can't
> >> be sent?
> >
> >Why?  All you need to send Hellos is a destination address and an
> >interface.  If you have a point to point link, then you know your
> >neighbor's address - either from configuration or from routing
> >protocols.
> >
> >If the interface is a broadcast-type interface, then you 
> would obviously
> >need some other mechanism to learn the neighbor address(es).  If you
> >know them, however, there is no reason why you couldn't send 
> Hellos as
> >well.
> >
> >-- David
> 
> Then if have to have "some other mechanism" learn them, then 
> we are out of
> the RSVP scope. 
> 
> I think we got the initial point accross. ;-]
> 
> Bill
>