A Transport Protocol for Multimedia Transmission in Overlay Networks Byung-Seok Kang, Kwangcheol Shin, In-Young Ko Department of Computer Science, Korea Advanced Institute of Science and Technology (KAIST), 335 Gwahangno, Yuseong-gu, Daejeon, 305-701 Korea {byungseok, kcshin2005, iko}@kaist.ac.kr Pyoung-Yun Kim Department of Electronics Engineering, Korea University, 5-ga, Anam-dong, Sungbuk-gu, Seoul, 136-701 Korea py81.kim@gmail.com Abstract— Overlay networks are opening up new ways for Internet usability, mainly by adding new services that are not available or cannot be implemented in the current Internet. Overlay networks use application level switches of intermediate nodes. However, the designs of overlay network are based on store-and-forward principles maintained by split TCP connections of hop-by-hop approaches and this can increase delay of packet delivery and cause unreliability of packet transmissions due to buffer overflow of the application level switch of each intermediate node, which has a limited buffer space. Also, hop-by-hop approaches for overlay networks cannot support end-to-end semantics. Thus, this study proposes a new Overlay Transport Protocol for overlay networks by considering management of end-to-end connection and congestion and flow control between source and destination nodes. By using a prototype implementation via simulation with MATLAB, We validate our analytical findings and evaluate the performance of proposed protocol which can provide end-to-end reliable data transfers, and also improve throughput in comparison with a network model where Overlay Transport Protocol layer is not available. Keywords- Overlay Transport Protocol; Overlay Network; Congestion and Flow Control I. INTRODUCTION With the increasing needs for multimedia applications, a number of services are being deployed on Internet for the applications. Over the last few decades, TCP has served as the dominant protocol for various network services deployed in the current Internet. However, TCP has difficulties to effectively support for various multimedia applications due to its fundamental limits. Overlay networks [7, 8] are opening new ways to Internet usability, mainly by adding new improved services, which are not yet available or cannot be implemented on the current Internet, such as Peer-to-Peer communication, Content Delivery Network, Application Level Multicast, and Voice over IP. All of these services use a virtual network over existing physical network. An overlay network has a topology of nodes connected at the application level layer and consequently, the usage of overlay networks may come at a price usually in added latency, which is incurred due to longer paths created by overlay routing and due to process the messages in the application level of every overlay node on the path. In overlay networks, the end-to-end connection must be split into hop-by-hop TCP connections, and packets, which are sent from a source node to a destination node, should go up to the application switch layers of all intermediate nodes along the overlay path. This split connection breaks TCP end-to-end semantics and as a result, congestion and flow controls are performed not between the two end nodes but between two adjacent nodes. These hop- by-hop split connections cause unreliability along the overlay path and also topology complexity are increased because many nodes and functions are required to support the services. This work proposes a new Overlay Transport Protocol (OTP) for overlay networks which performs congestion and flow controls between source and destination nodes to manage reliable end-to-end connection. We evaluate the performance of OTP by using a prototype implementation via simulation with MATLAB. Through the simulation, OTP shows that it can provide end-to-end reliable data transfers, and also improve throughput in comparison with a network model where OTP layer is not available. This paper is organized as follows. In Section II, previous related works are discussed. We introduce the Overlay Transport Protocol and present experimental results in Sections III and IV, respectively. Finally, Section V concludes the paper. II. RELATED WORK Protocols for end-to-end congestion and flow controls can be divided into three categorized as hop-by-hop, end-to-end and application level switch approaches. In this section, previous works of those three categories are explained. A. Hop-by-hop Approach Hop-by-hop approaches for congestion and flow control in overlay networks have been researched over the last few years. H. Pucha et al. [1] proposed an overlay TCP that splits the direct TCP connection between any two hosts into a multiple- overlay-hop path. By splitting the direct TCP connection into multiple connections, the RTT of each overlay hop is lower than the direct RTT between two end hosts of the direct TCP connection. Then, the hop throughput of each overlay will be higher than the direct throughput between two end hosts. But 2010 24th IEEE International Conference on Advanced Information Networking and Applications 1550-445X/10 $26.00 © 2010 IEEE DOI 10.1109/AINA.2010.8 669