Evaluating and improving the deployability of Multipath TCP

More than forty years since the creation of the first ARPANet nodes, the Internet continues to evolve. The recent decade has seen the introduction of new types of devices that are multi-connected: smartphones and tablets. These devices often have both a WiFi and a mobile data connection. Unfortunately, the TCP/IP protocols were designed 30 years ago when devices only used a single interface. As smart devices are multi-connected, users expect that their applications can seamlessly handover from WiFi to 3G. This mismatch between what the protocols provide and what the user expects therefore impacts the user experience. Recently, a new extension to TCP, called Multipath TCP (MPTCP), has been proposed and standardized. MPTCP allows a single connection to use multiple interfaces at the same time actually adding the missing mobility feature to TCP. This extension, as every new protocol/extension, suffers from the chicken-and-egg problem. To use MPTCP both clients and servers have to support it. This PhD thesis focuses on the deployability of MPTCP and evaluates its benefits in different network environments. First we evaluate the benefits of using MPTCP on a smartphone as to show that indeed MPTCP improves the end user experience. Second we evaluate if MPTCP can be used on the Internet by identifying middleboxes that could prevent it from being used. We then propose to solve the chicken-and-egg problem by deploying protocol converters on the Internet that translate TCP to MPTCP. This converter gives incentive to smartphone vendors to implement MPTCP on their devices and use it to improve the customer experience. Finally, we propose a new load balancing technique that improves the link usage in data centers.