... Networks¹

This research was supported in part by NSF-ARI grant No. 9601602.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

...š²

Corresponding author.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... primarily³

An all-optical WRN consists of a plethora of other devices as well, but none of these are pertinent from the perspective of the current discussion.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... connected⁴

A node failure can be modeled as a multiple link failure.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... reliable⁵

The DCN can be protected using 1+1 protection switching as discussed in [4].

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... headend⁶

The node immediately upstream to the failed link.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... same⁷

Here $t_n$ still represents DCN delay, but unlike serial OBS activation, $\forall n \in \mathcal{P} : t_{n+1} > t_{n}$ does not necessarily hold for internetworks in this case.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... approximately⁸

For each OTN node, the DCN node with the smallest Euclidian distance from it was chosen. However, this still does not guarantee the best geographical superimposition of the DCN on the OTN with all distances minimized.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... CAIDA⁹

http://www.caida.org.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... two-fold¹⁰

Since the DCN is a transit-stub graph, the nodes in the DCN increased by a factor of $Z = 4 + 4 \cdot \frac{s}{s+1}$ where $s$ is the number of nodes in a stub.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.

... 0.23¹¹

A relatively higher edge-probability is required in order to guarantee 2-node connectedness using the gt-itm package.

.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.