Abstract:Recently, with the development of new technologies like virtual reality, Internet of Things, cloud computing, and so on, the user demand on network bandwidth increases sharply, and it is hard to meet user bandwidth demand by making use of single access technology. To solve the contradiction between users' increasing bandwidth demand and limited frequency resources, Internet end-to-end multipath transmission technologies are designed. Internet end-to-end multipath transmission protocols, like MPTCP (multipath TCP), mainly work at transport layer, and they can make use of existing multiple network cards (for example WiFi and 4G) to do concurrent transmission, the total transmission bandwidth and the adaptability to network variation are improved. Since the subflows of MPTCP can realize end-to-end reliable in order delivery based on TCP and its optimizations, research works related to multipath transmission focus on the intelligent coorperations among subflows, the coorperations mainly include subflow selection, data scheduling, coupled congestion control, and so on. Nevertheless, as the estimation of network status at transport layer may not reflect current link state accurately due to the variation of link and physical layer, heterogeneous network interfaces have different resource allocation schemes, different subflows may have partially overlapped paths, the upper application layer have different characteristics in deadline, frame importance, distortion rate, and so on. These will affect the performance of the intelligent coorpertaions among subflows, the advantage of multipath transfer can not be fully exploited by only making use of traditional transport layer information, the information at other layers should also be utilized. So in recent years, many research papers start to make use of physical, link, network, and application layer information to improve multipath transmission performance efficiently with cross layer cooperated optimization. This paper compares the research of multipath transfer optimization with cross layer information, analyzes the relationship between multipath transfer and the function of different layer, and the outlook of future research trends is provided in the end.