Performance evaluation of MPLS and IP on an IXP1200 network processor

One of the most suitable protocols for traffic engineering is multi-protocol label switching (MPLS). On the one hand, a look-up in a switching table is certainly less complex and time-consuming than a corresponding routing table lookup in an IP router as long as no specialized hardware such as a Content Addressable Memory is deployed. On the other hand, MPLS introduces an additional shim header that increases packet transmission delay. The question this paper intends to answer is thus if the common statement that MPLS generally allows for faster packet forwarding than IP is true. Therefore, an extensive performance evaluation based on an implementation on standard network processors is performed. It is shown that packet forwarding is only faster with MPLS when the packet size remains below certain bound, i.e. it is demonstrated that the common statement that MPLS always allows for faster packet forwarding does not even hold on standard hardware.