Also agree in principle with Vishal's path computation
hierarchy. When dealing with different granularity
signal types, e.g., from 1 or 2Mbps MPLS LSPs all the
way to lambda or waveband switched GMPLS LSPs
different methods are generally called for. For
example, a distributed PCS approach would initially be
overkill, versus a centralized approach (not that many
LSPs to get set up that often) while if one is dealing
with lower rate MLPS LSPs a distributed PCS approach
(hopefully) would be more appropriate.