The MPLS WG Archive

["Indra.Widjaja" <Indra.Widjaja@tddny.fujitsu.com>]

Don:

 

Maybe we're looking at different problems.
My understanding of the problem is that you would like to
compute n working LSPs on a network using an offline tool,
and each working LSP is to be protected by another protection
LSP using a common option; e.g., 1:1 or 1+1 (it may be
expanded to include shared protection).
The question is that would it be better for the protection
LSPs to be computed offline or online? 

If you do it that way, offline may be reasonable. The question is:  Is that the majority

of your LSPs or just a selected few ?  A selected few is O.K. but for the rest 

The number of LSPs computed offline depends on the network environment.
In particular,
if the traffic demand for a given node pair is predictable for a relatively
long duration,
then there is a role for computing such an LSP offline. On the other
hand, if the traffic
demand cannot be predicted, then one has to resort to an online system
only.
It is generally not worth doing offline computation if only a few LSPs
are
computed offline since the major advantage of an offline system 
is better network
efficiency. This is because a few LSPs computed offline would mostly
not result in
a significant overall improvement. Of course, there are specific cases
which this
may not be true.

 

online scales better and is more manageable.  What I was talking about was 

restoration of LSPs that may take a little longer but is much less expensive.

So you're refering to protection LSPs that are established "on-demand"
after
failure notification is received, in which case an online system is
the only option.

indra