Transport protocols: limitations, evolution obstacles and solutions for an actual deployment in the Internet

International audienceThe Transport layer, designed for old networking contexts and now obsolete applications requirements, is inefficient. This paper discusses the reasons behind this inefficiency and the obstacles to the evolution of Transport protocols. The discussion is then extended to derive new requirements for the Transport layer, both functional and architectural, in order to ensure optimal performances in all current and future contexts. To meet these new requirements, a novel architectural design of the Inter-net Transport layer is proposed following a service-oriented and a component-based approach. The proposed solution allows for optimization of the Transport service performance, facilitation of its utilization, and is aimed at allowing the integration of new services as needed. 1. Introduction The Transport layer is one of the most important layers of the Internet communication protocol stack. Located between the application and the network layers, this layer is expected to take into account both the application feature/requirements and the underlying network capabilities/constraints in order to provide the best end-to-end communication service, matching as much as possible the application requirements, still taking into account the opportunities and limitations of the underlying network. In the last few decades, several new applications and networking technologies have emerged with very different requirements and characteristics, making initial Transport-level protocols (typically TCP and UDP in the Internet) no longer adapted. As a consequence , a lot of new Transport protocols and mechanisms have been proposed over the last 20 years in order to enhance both the Transport services offered to applications, and to optimize the usage of the different network technologies. Unfortunately, the factual observation is that all these new proposals are either not integrated in the actual operating systems (Windows, Linux, etc.), or hardly used in practice by the application developers, which in majority continue to use suboptimal TCP and UDP solutions. Since it is proved that these new propositions are widely more efficient than the actual used protocols, the main goals of this work are to identify the deploymen