A Highly Parallelizable Algorithm for Routing With Automatic Tunneling

Most current routing protocols are based on path computation algorithms in graphs (e.g., Dijkstra, Bellman-Ford, etc.). These algorithms have been studied for a long time and are very well understood, both in a centralized and distributed context, as long as they are applied to a network having a si...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:IFIP Networking Conference S. 1 - 9
Hauptverfasser: Mouhoub, Noureddine, Lamali, Mohamed Lamine, Magoni, Damien
Format: Tagungsbericht
Sprache:Englisch
Veröffentlicht: IFIP 13.06.2022
Schlagworte:
ISSN:1861-2288
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Most current routing protocols are based on path computation algorithms in graphs (e.g., Dijkstra, Bellman-Ford, etc.). These algorithms have been studied for a long time and are very well understood, both in a centralized and distributed context, as long as they are applied to a network having a single communication protocol. The problem becomes more complex in the multi-protocol case, where there is a possibility of encapsulation of some network protocols into others, therefore inducing nested tunnels. The classic algorithms cited above no longer work in this case because they cannot manage the protocol encapsulations and the corresponding protocol stacks. In this work, we propose a highly parallelizable algorithm that takes into account protocol encapsulations as well as protocol conversions in order to compute shortest paths in a multi-protocol network. To achieve this computation efficiently, we study the transitive closure between sub-paths (i.e., the concatenation of two sub-paths to obtain a longer one) in the case where each sub-path induces a protocol stack, and thus tunnels. Leveraging on Software-Defined Networks with a controller having a highly parallel architecture enables us to compute the routing tables of all nodes in a very efficient way. Experimentation results on both random and realistic topologies show that our algorithm outperforms the previous solutions proposed in the literature.
ISSN:1861-2288
DOI:10.23919/IFIPNetworking55013.2022.9829784