The MPLS WG Archive

["Darek Skalecki" <dareks@nortelnetworks.com>]

While we are on the topic of setup timers, is there a timer in RSVP-TE which times
how long a PSB exists without a corresponding RSB. That is, should RSVP-TE start a
timer after creating a PSB upon processing of a PATH message. If a corresponding
RSB is not created within specified time, i.e. RESV message doesn't arrive within
that time,  then the PSB is declared "uncompleted" and potentially destroyed,
maybe even PATH_TEAR generated upstream.

Such a timer would be useful in particular since a while back it was indicated on
the mailing list that bandwidth reservations for RSVP-TE are done on the
PATH message as opposed to on the RESV message as in classical RSVP. Such timer
would thus prevent PSBs from unnecessarily holding onto bandwidth for long periods
of time.

I agree with Yangguang that a concern about setup timer for both CR-LDP and
RSVP-TE is valid to ensure that once LABEL_REQUEST or PATH messages are launched
then LABEL_MAPPING or RESV (or error) messages arrive within a predefined time or
else the LSP establishment is aborted.

Darek


Yangguang Xu wrote:

> Please see further questions below:
>
> Chatur sharp wrote:
> >
> > It runs on TCP, which provides this garuntee. Incase
> > response is not received the TCP times out. Much like
> > T301 etc.
>
> Anything can happens. What happens if TCP session is OK, but CR-LDP packets are
> dropped some place, for example, buggy CR-LDP code? Are we sure that TCP is
> enough? Do we have any requirement for CR-LSP setup time? Any requirements from
> service provider?
>
> > A couple of points to remember are :
> > 1. Q.2931 is a UNI i/f where message from
> >    one user (caller) are carried by the network
> >    ( through possibly multiple hps ) to the second
> >    user ( receiver).
> >
> >    LDP is not like that it is really Peer to Peer.
> >    So protection for individual calls is not really
> >    required. ( Although it would be desirable for
> >    some other purpose )
> >
> > 2. Typically TELCO protocols have more reliability
> >    redundancy built into the stack than datacom world.
> >
> >
>
> First, I should say PNNI.
>
> Second, MPLS is mainly defined for core data network, where reliability is as
> important as any traditional Telco network. PNNI or Q.2931 is a little bit too
> complicated, however, I think timer is still needed for CR-LSP setup.
>
> Yangguang