Revisiting the IETF multipath extensions on transport layer

Load sharing on the transport layer of the OSI reference model is an important topic in the IETF standardization. This approach is also supported by the industry to optimize the use of the resources in a network like the Internet. After many trials, two basic sets of mechanisms and functionalities on the transport layer have been proposed by the IETF to achieve load sharing. These basic sets extend the protocol mechanisms that were originally designed for the use in singlepath dominated networks and represent only a first step to introduce a real end-to-end multipath transfer on the Internet. These first basic sets must be investigated and improved for the next steps. The Transmission Control Protocol (TCP) and the Stream Control Transmission Protocol (SCTP) provide the basis for the two IETF end-to-end multipath extensions. Both singlepath transport protocols have a different historical background but similar goals. These can be characterized by a reliable, connection-oriented and ordered data transport. However, initial experiments with the IETF multipath extensions in real networks show unexpected and in some cases clearly inadequate results. It is becoming rather apparent that the singlepath transport protocol specifications with their singlepath goals have a significant impact on the effectiveness of the load sharing mechanism and, furthermore, that the severity of the influence depends on the topology. The new mechanisms for multipath transfer include, in particular, an extended “path management” and “scheduling” task. The mechanisms addressing the path management organize the new, alternative paths and the scheduling mechanisms support their effective use. For both protocol extensions of TCP and SCTP, an interaction can be identified between the new load sharing mechanisms and the existing specifications for singlepath transfer. This thesis systematically identifies the impact factors of the singlepath specifications on the new load sharing mechanisms and demonstrates their effects. In addition to the focus on the optimal use, the fair distribution of resources across all connections must be taken into account in the IETF standardization process. This so-called “fairness” discussion is mandatory for a transport protocol in the IETF context and has a direct impact on the overall system performance. Furthermore, this thesis discusses the currently implemented load sharing extensions and analyzes their weaknesses. Moreover, in this work new design approaches are developed to decrease the impact.