Experimental Evaluation

Experimental measurements were performed in order to study the effect of each of the activation schemes on the restoration time. An experimental run consisted of measuring the average restoration time using the estimates calculated above for each of the four activation schemes for a given OTN and DCN network pair. Table 1 shows the typical values of various parameters used in the calculation of restoration time as suggested in [4]. The Georgia Tech Internet Topology Modeler (gt-itm) was used in order to generate pure-random [11], undirected OTN and DCN graphs. The OTN graphs were generated using the Flat Random Graph model provided by the gt-itm whereas the Transit-Stub Graph model was used to represent DCN structure [11,12]. The gt-itm package offers ways to control the network topology model, the number of nodes, the edge probability, the node-link distribution and the geographical network size for the graphs generated. In order to be representative of real networks, the geographical sizes of the OTN and the DCN were always maintained equal. A mapping was developed from each node in the OTN to a unique node in the DCN to model its controller. Using the geographical position information provided by gt-itm for each node in a graph, the mapping function ensured that an OTN node was always mapped to the approximately⁸nearest DCN node. Furthermore, the lengths of edges between a pair of OTN nodes were then set to the distance between their controlling DCN nodes to ensure a sound model. The OTN generated was always much smaller than the DCN in node-count (except in Experiment 3). Each data point was obtained by running measurements on sets of at least ten different randomly generated OTN/DCN pairs. Only 2-node connected graphs were used as OTNs for any experiment. All $\frac{(p-1)\cdot p}{2}$ possible shortest paths in an $p$-node OTN were considered to be primary paths. All $\frac{(p-1)\cdot p}{2}$ possible node- and link-disjoint second-shortest paths in an $p$-node OTN were considered to be secondary or alternate paths. Each data point represents the mean of the time to restore traffic from a primary path (shortest path) to its node- and link-disjoint secondary (second-shortest) path. The Stanford GraphBase [13] library was used for writing routines to perform measurements. The Walrus Graph Visualization tool from CAIDA⁹  was used for visualizing graphs.