A framework for preparing experimental evaluation of rerouting mechanisms

Abstract

Among the various QoS schemes that have already been proposed, rerouting critical flows is the most promising for its inherent ability to interface easily with management systems. To enhance the responsiveness, the operations performed by rerouting mechanisms should be done pro-actively. A pro-active network management using rerouting mechanisms was proposed by Correia et al., (2003). Recently, the proposed approach was selected to be deployed in an experimental fiber-optic gigabit backbone (GIGA project - sponsored by the Brazilian Research Network - RNP). The fundamental problem is how to specify critical factors and to develop a consistent set of tests for a complete field evaluation of the implementation. The constraints associated with this process are reduced deployment time and restricted availability of the network infrastructure for testing. This paper discusses a framework for deploying the rerouting mechanisms proposed by Correia et al. (2003) using DOE (design of experiments) analysis (Phadke). It is shown how rapidly some conclusions about performance, stability and interdependence upon the selection of design parameters can be taken through the use of DOE. Basically, MatLab models have been used to support the DOE analysis. This paper shows how the use of simple two and three-level factorial experimental designs can rapidly increase the designer's knowledge about the behavior of the mechanisms being studied.