The Routing Over Large Clouds Mailing List Archive by date

[Curtis Villamizar <curtis@ans.net>]

In message <199510260821.KAA02422@lohi.dat.tele.fi>, Juha Heinanen writes:
>    If the application specifies QoS and traffic characteristics and
>    indirectly controls SVC management, then we are in agreement.  Then it
>    doesn't matter if the box it is running on sets up a VC or if it sits
>    on an ethernet and the next hop sets up a VC.  Are we in agreement on
>    this?
> 
> this is what i have never understod about rsvp.  how can a host sitting
> on a shared media, like ethernet, make any reservation whatsoever.  so
> isn't it a requirement for rsvp to work that the hosts are directly
> connected to such a link layer network, eg. atm, that supports
> reservation requests?  what does it help if all the routers in the whole
> internet are rsvp capable if the last hop isn't?
> 
> -- juha

The application states a traffic characterization and requirement.  If
the ethernet is thought to be uncongested, the traffic must then make
a one hop "leap of faith".  It can then go through a congested T1
(assuming RSVP is supported and the reservation is accepted) without
further loss, or traverse an NBMA.  If the ethernet is not
uncongested, things don't work very well.  If we are talking about a
fairly dedicated ethernet segment, or somewhat traffic isolated by
10baseX hub, then the "uncongested" assumption may be close enough.

Curtis