RSVP performance evaluation using multi-objective evolutionary optimisation

Abstract

The proposed uses of the resource reservation protocol (RSVP) now extend beyond reserving resources in Internet protocol (IP) networks to being a generic signaling protocol for generalised multi-protocol label switching (GMPLS). In any implementation of RSVP, there are a number of discretionary timing parameters, the values of which affect the efficacy of RSVP in establishing and maintaining reservations/connections. This work frames the interactions between key RSVP timing parameters and performance metrics as a multi-objective optimisation problem which, due to its intractable nature, is tackled using a reputable multi-objective evolutionary algorithm. It is shown that this approach is a feasible means of exploring many of the innate tradeoffs in soft-state protocols such as RSVP. This approach facilitates an extensive comparison of a number of variants of RSVP: standard RSVP, RSVP featuring the recently standardised retransmission algorithm and two subsequent variants of this algorithm, supporting the asymmetric delivery of RSVP control messages. These RSVP variants are compared in terms of multiple performance metrics under a number of different exemplar network conditions, giving insight into their relative merits. Furthermore, the relative significance of the different timing parameters is investigated and their most expedient values determined.