Adaptive routing approach in networks with MPLS technology

Roman Ganopolskyi¹, post graduate student

1 Computer System Department, Faculty of Information and Computing Technique, National Technical University of Ukraine “Kyiv Polytechnic Institute”, Address: 37, Prospect Permohy, 03056, Kyiv-56, Ukraine, e-mail: celenia_rga@hotmail.com

One of the main criteria of now-days networks user satisfaction is Quality of Service, QoS. However, TCP/IP technology in its present state cannot achieve these requirements due to using of inefficient OSPF routing, which causes congestions and connection rejects while traffic load increases. Multi Protocol Label Switching technology, MPLS, is an actual solution for QoS problem. It offers traffic classification, explicit routing which allows efficient load balancing which leads to obtaining needed quality of service.

Different routing schemes for MPLS were introduced. As researches show the most efficient are dynamical routing algorithms, which use actual network state. Also, statistic information about connection requests in network is useful information as well, because it allows predicting future state of network. Some routing algorithms, so-called online algorithms, allow rerouting (deviating) traffic flow while it is being routed (VFD). Others, offline, don’t provide such an ability, which significantly reduces computational complexity, but loses flexibility.

Performance of routing algorithms can be measured using such metrics as connection reject probability for set some network load, computational complexity, memory requirements, service information percent in total traffic.

There are several mathematical models to evaluate performance bounds, particularly rejection probability. The most precise are Ideal routing model (IR) and Min-Cut model based on multi-class Erland formula. The closer modeling result of routing algorithm to model the more efficient algorithm is.

Among offline algorithm the Minimum Interference Routing Algorithms, MIRA, are known to be efficient. They provides scheme, where system try not to put path across other paths (for path not to interfere with others). However, these routing schemes don’t use statistics and though cannot predict future connections to be effective set up for receiving connections from most frequent directions.

Researches show that modification of MIRA with least computational complexity and taking into account statistics allows achieving good performance in comparison with more complex schemes while saving time for calculation.