## Abstract

Physical layer impairments (PLIs) need to be considered in the routing and
wavelength assignment (RWA) process of all-optical networks to ensure the
provisioning of good quality optical connections
(i.e., *lightpaths*). A convenient way to model the impact of
PLIs on the signal quality is to use the so-called *Q*-factor. In
a dynamic provisioning environment, impairment-aware RWA (IA-RWA) algorithms
include *Q*-factor evaluation in their on-line decisions on
whether to accept a connection request or not. The *Q*-factor can
be computed in either an *approximated* or an
*exact* way. IA-RWA algorithms using an approximated
*Q*-factor estimation (i.e., worst case) can be very fast and
allow for a short setup delay. However, connection request blocking can be
unnecessarily high because of the worst-case assumption for the
*Q*-factor parameters. In contrast, an exact
*Q*-factor computation results in a better blocking
performance at the expense of a longer setup delay, mainly due to the time spent
for the *Q*-factor computation itself. Moreover, an exact
*Q*-factor approach requires extensions of the generalized
multi-protocol label switching suite. To overcome these problems, we propose a
statistical approach for fast impairment-aware RWA (SAFIR) computation. The
evaluation results reveal that SAFIR improves the blocking probability
performance compared to the worst-case scenario without adding extra
computational complexity and, consequently, without increasing the connection
setup delay.

© 2012 OSA

Full Article | PDF Article