18. Ayudantes:
3.6. Proyecto Casa Sandia
3.6.4. Descripción del Proyecto Casa Sandia
This section discusses three dimensioning techniques that estimate the traf- fic loads to each border node (i.e., pn,s,p). This depends on the amount
of information shared across domain borders and the inter-domain routing agreement.
Three dynamic routing algorithms are used here. The first two are the same as the algorithms found in [50]. In source-initiated concatenated shortest path routing (CSR), the requesting node chooses the closest border node and uses the shortest path. Starting from this border node, the path in the other domain is selected. No information other than the desti- nation is shared between border nodes. If the downstream domain cannot find a path to the destination through the selected border or the inter-border link is full, the call is rejected.1 End-to-end global shortest path rout-
ing (E2E) chooses the border nodes that result in available shortest global paths. Network domains do not have to disclose detailed topology and link load information to other domains. For each connection request, they need broadcast only the available (reachable) shortest path distance from source to each border for the source domain, from each border to all other borders for the intermediate domains, and from each border to the destination for the destination domain. Then, the global shortest path can be found by choos- ing the combination that yields the shortest length global path. If a domain has no available paths or if two domains cannot join paths due to resource
unavailability, the call is rejected. Equation 7.5 shows the computation of shortest paths between connected border pairs in two-domain networks. The border pair is picked with minimum lE2E
b,b′ . Normalized global shortest
path routing (nE2E) is similar to E2E, but each reported distance is normalized by the average TSL of each respective domain before being ad- vertised and summed. Equation 7.6 shows the computation of nE2E path length. The pair of border nodes is picked with minimum normalized dis- tance lnE2E
b,b′ . nE2E is important because E2E favors the border node selection
for the larger network when two domains are of different sizes: since the the larger domain is more likely to have longer paths to the border, its shortest path length can dominate the length of the global shortest path and force the smaller domain to pick unfavorable paths.
For example, assume available SPF lengths from a source node to three borders are 5, 7, and 11 hops in the larger domain A (average TSL=7). In the smaller domain B (average TSL=2), three corresponding borders to the destination node can be 3, 2, and 1 hops away. In E2E, the path with 5 hops in A and 3 hops in B is chosen. In nE2E, the path with 7 hops in A and 2 hops in B is chosen. Although the path picked by nE2E is longer, it is more fair and yields better overall performance for both internal and external traffic on the network, if the network is also dimensioned for nE2E.
lE2Eb,b′ = T SL(n,b)+ T SL(s,b′) (7.5) lnE2Eb,b′ = |R|T SL(n,b) P i∈RT SLi +P|S|T SL(s,b′) i∈S×ST SLi (7.6) I now describe the computation of pn,s,b in three dimensioning algorithms.
Once the distribution is computed, I can use the basic dimensioning algo- rithm (Algorithm 3.2) to dimension each network separately. Independent shortest path dimensioning (IS) is used for two networks that share only node information.2 In this case, an internal node has no idea which border
node(s) external calls will come from. Therefore, external traffic is split uni- formly across all border nodes, in this case, pn,s,b = |N1′|. Global shortest
path dimensioning (GS) allocates wavelength resources using least-cost
2Node information is the minimal amount of information needed to support full
bandwidth-on-demand services; no request is ever made without knowing which desti- nation node to connect to.
Algorithm 7.1: Computation of pn,s,b for GS/NS.
1 foreach inter-domain connection pair (n, s) do
2 foreach border node pair (b, b′) from each domain do
3 Compute the total path length lE2E
(b,b′) (or lb,bnE2E′ for NS);
4 Identify the subset of borders B⊆ N′ of minimal total path length;
5 foreach shortest path border nodes b∈ N′ do
6 pn,s,b= 1
|B| for b∈ B, pn,s,b= 0 for b /∈ B;
routes crossing two domains (assuming E2E). This approach is used when two networks are willing to share path lengths at the border node (the ap- propriate E-NNI interface) to pick the global shortest path. The traffic rate to each border node is then weighted by the likelihood that each border node will be chosen using E2E. Algorithm 7.1 shows the computation of pn,s,b for
each inter-domain connection pair (n, s). Normalized shortest path di- mensioning (NS) is similar to GS except for the metric used to compute the concatenated shortest path length. Algorithm 7.1 shows the procedure to compute pn,s,b for NS, assuming the use of nE2E.