Bandwidth Aggregation with SCTP Antonios Argyriou and Vijay Madisetti School of Electrical & Computer Engineering Georgia Institute of Technology Atlanta, GA 30332, USA Email: {anargyr, vkm}@ece.gatech.edu Abstract— In this paper we present a number of modifications to the recently adopted by IETF Stream Control Transmission Protocol (SCTP), that allow bandwidth aggregation over the multiple interfaces of a host. We show that it is possible to implement a number of algorithms for bandwidth aggregation, with only a small number of modifications to the base SCTP protocol. Our simulation results clearly depict the efficiency of our approach in terms of bandwidth utilization. Furthermore, we implement and evaluate a mechanism for identifying bottlenecks that are shared by flows from the same aggregate connection. Our purpose is to show that SCTP is a good candidate for building a practical protocol for bandwidth aggregation that is fair and supportive of TCP. I. I NTRODUCTION Bandwidth was always an immediate metric for the user perceived QoS. That is why end-user terminals are equipped today with a number network interfaces (LAN/WLAN) which in the near future will probably belong to various access technologies. Ideally, a user would like to use all the available aggregate bandwidth provided by the available interfaces. Moreover, service providers require better resource utiliza- tion for their equipment and one way to achieve this is by accommodating more users to the same infrastructure. Using concurrently multiple technology interfaces (aka multiple ac- cess networks) can be a source of increased revenue for them. Thus, the problem that formulates is how to use all theses interfaces in order to obtain the maximum possible bandwidth in a fashion friendly for the Internet and the dominant TCP traffic. In this paper we propose a mechanism for efficient band- width aggregation (or load-sharing) across the multiple inter- faces of a multi-homed host. The proposed approach con- sists of a number of modifications to the Stream Control Transmission Protocol (SCTP), recently adopted by IETF [1]. This protocol includes a number of novel features, including support for multi-homing. However, the current SCTP speci- fication does not implement bandwidth aggregation due to the difficulty of the problem. We believe that due to SCTP’s novel features, it has the potential of providing a robust protocol for practical bandwidth aggregation over the Internet. Our paper tries to give a first direction towards a practical bandwidth- aggregation solution based on SCTP that will be able to operate smoothly over the current Internet. Additionally, we evaluate a mechanism for solving the most important problem of a bandwidth aggregation system: TCP-friendliness/fairness of the aggregate connection even when the sub-flows from the same host converge to the same bottleneck router. Overall, our system functionality can be summarized into three key points: 1) shared bottleneck identification 2) recovery with a modified congestion control algorithm and 3) tackling out-of- order delivery of packets due to path asymmetries. The rest of this paper is organized as follows: In sec- tion II we give an overview of work related to bandwidth aggregation and load-sharing. Due to the recent emergence of SCTP, we provide a brief overview of the protocol in section III. Section IV provides motivation for our work and analyzes in detail the problems we are dealing with. The next section describes in detail our algorithms and the proposed modifications to SCTP. The algorithm for shared bottleneck identification and recovery is presented in section VI. Finally we present simulation results in section VII, while section VIII concludes the paper. II. RELATED WORK A number of application layer techniques for bandwidth aggregation have been presented [2], [3], [4]. The common practice at this level is to establish multiple TCP connec- tions (each one mapped to a different interface) and then stripe application data according to the available bandwidth of each link. Bandwidth estimation is performed either pas- sively (RT T measurements) or actively using probes. These approaches obviously tradeoff accuracy for wasted bandwidth. Another major problem is the need of large re-sequencing buffers in the case of significant mismatch in the available bandwidth of each link [5]. Recently in [4], was presented a application layer bandwidth aggregation system, that was built based on the analytical results presented in earlier work [6]. Their system performs well in identifying shared bottlenecks, except the case of sources that exhibit bursty traffic patterns. Moving down to the protocol stack, we find the first trans- port layer bandwidth aggregation technique being reported in [7], in the form of a specialized protocol named RMTP. This protocol is based in the accurate bandwidth estimation of each path for proper data striping. However, since the bandwidth estimation accuracy depends largely on the frequency of probes sent to each path, significant bandwidth is given out for probing. Recently, Hsieh [5] proposed a modified version of TCP, called P-TCP, for bandwidth aggregation in mobile multi- homed hosts. However, the assumption of sub-flows that do not share a bottleneck limits the applicability of the protocol in the current Internet. Moreover, the additional requirement GLOBECOM 2003 - 3716 - 0-7803-7974-8/03/$17.00 © 2003 IEEE