Evaluation and Prototyping of Multipath Protocol Mechanisms

Multipath transport is a promising new Internet par adigm. Most Internet applications use the Transmission Con trol Protocol (TCP), which cannot efficiently use multip le available paths between the connection endpoints. As a conseq uence, there is a significant potential for new protocol m echanisms that overcome this limitation of the current Intern t. There are several situations in which data transpor t over multiple paths would make sense: End users access t h Internet more and more from mobile devices that have multipl e network interfaces, which could be used in parallel to use their aggregated bandwidth. Also, data centers are typica lly connected to the Internet by several uplinks in ord er to improve the reliability and to enable load balancing (see F ig. 1). The latter use case is particularly important for netwo rk-based computing applications that are realized in the clo ud, that have high performance and reliability requirements, and that are the focus of the G-Lab NETCOMP project [1]. Multipath transport is not an entirely new concept and has been researched before. Still, there are new efforts to develop multipath transport protocols that could be used at Internet scale [2]. The Internet Engineering Task Force (IET F) has started corresponding standardization activities [3 ]. This abstract describes work-in-progress that devel ops, evaluates, and prototypes multipath protocols for N ETCOMP applications. In Section II we first survey and cla ssify multipath solutions and identify open issues that r equire further work. Two examples are the design of interfaces tow ards applications [4], and the cross-layer interaction b etween transport and routing. Section III presents the cur rent status of our prototyping activities for new multipath protoc ol mechanisms. Section IV briefly summarizes some less ons learnt from initial measurements. Finally, this doc ument concludes with a summary and an outlook.