Best-effort traffic engineering in multiprotocol label switched networks

This work addresses the problem of engineering best-effort traffic in Internet Protocol (IP) networks. Traffic engineering, in general, consists of routing traffic flow in order to achieve performance gains or meet policy objectives. This work focuses on engineering traffic to increase performance. The recent resurgence of interest in IP traffic engineering is due to the recent introduction of Multi-protocol Label Switching (MPLS). MPLS provides the connection-oriented features required to route traffic over arbitrary paths; it makes traffic engineering in UP networks possible. This work presents a general traffic engineering framework and four specific implementation algorithms. The framework has two salient features: (1) It uses MPLS to encapsulate source-destination aggregate flow within a Label-Switched Path (LSP), with multiple LSPs per source-destination. (2) The framework is state-independent; only the topology is used to determine traffic routing. The framework was evaluated using both randomly generated topologies and a production Internet Service Provider (ISP) backbone. On average, performance improvements on the order of 5%–15% over Shortest Path (SPF) routing were observed.