Evaluation of the XL Routing Algorithm in Multiple Failure Conditions Nguyen Cao Julie Morris Khang Pham

Introduction Background material for XL algorithm Problem Description Approach Taken Results

Background The XL routing algorithm is a new link-state routing algorithm. Introduces three criteria to determine update propagation; otherwise, link cost updates are suppressed.

Three Criteria S1: When the updates is a cost increase S2: When the link is used in the node’s shortest-path tree C1: When it improves the cost to any destination by more than a 1 + ε cost factor First two criteria ensure soundness, third ensures completeness and bounded stretch

XL’s Three Phases Phase I: Calculate preliminary views, internal and external Ensure local external view is up-to-date with neighbor’s copy of external view. Phase II: Update the forwarding table if necessary Phase III: Propagation decisions based on criteria

Problem Description The XL paper performs an evaluation of their own algorithm as compared to other routing algorithms. We aim to validate these results. We also evaluate the XL algorithm in multiple failure scenarios to determine real world viability. This stress-testing is intended to determine the suitability for application in real networks which may have large numbers of failures.

Problem Description, cont. To accurately assess the XL algorithm, several parameters must be monitored: Convergence time Optimality Number of updates sent To validate claims, convergence time and number of updates sent should be smaller while maintaining near optimality In multiple failure situations, convergence time should be reasonable and ensure correct re-routing of packets.

Approach Taken We have implemented the XL algorithm and simulated using OMNET++ OMNET++ allows us to describe what a node is and submodules for handling packets We created a Routing submodule to handle all traffic forwarding and an Algorithm submodule to handle packets delivered to the node indicating cost changes or view propagations The Algorithm submodule implements the three phases of XL

Network model

Link Cost and Failure Simulation To simulate dynamic network changes, we implemented an independent special node which randomly chooses a link to change. In high failure mode, the link changer will kill this link with a 50% probability The link changer notifies the nodes on each end of the link of the change through a special gate.

Topologies for Testing Implemented three small synthetic topologies described in the XL paper- Crown, Honey, and Quad Also implemented larger RocketFuel topology with 300+ nodes and 700+ links Model of actual internal AS topology in ISPs

Results Due to technical difficulties in ensuring our implementation of the algorithm is performing to specification, we are only able to provide preliminary results. Preliminary results from running on the Crown network in high failure mode: Nodes: 36 Convergence time: 172 secs with 2 link changes Total routing updates sent: 2726 The outline of simulations we will be running: Show convergence times and number of updates for networks in regular situations vs. the high failure mode. Assess whether high failure mode still produces a sound and complete network. Show convergence times and number of updates on smaller network topologies vs. the RocketFuel topology.

Future work Adjust the approximate epsilon dynamically in response to load. Deterministic routing in normal circumstances, approximate routing under heavy load. Improve the criteria especially the condition S1: always propagate bad news (cost increase) somehow still costly. Selectively propagate changes when link failure.

Demo

Thank you! Questions?