Demonstration on Fairness Among Heterogeneous TCP Variants Over 10Gbps High-speed Networks

Several high speed TCP variants are adopted by end users and therefore, heterogeneous congestion control has become the characteristic of newly emerging high-speed networks. In this demonstration, we show the fairness among heterogeneous TCP flows over CRON, a 10Gbps high-speed networking testbed. The significant throughput difference among heterogeneous TCP flows shows that heterogeneous TCP flows induce more unfairness than homogeneous ones. Moreover, fairness behavior changes as we change intermediate router parameters, such as queue management scheme, buffer size, etc. During the demonstration, audience will have an intuitive view on fairness behavior among heterogeneous TCP variants over 10Gbps high-speed networks. I. SCOPE OF THE DEMONSTRATION High-speed networks with capacity to deliver a data transfer rate in excess of 10Gbps have been developed to serve the demand of end users. Among all the characteristics of highspeed networks, fairness among its end users has been one of the most important. A recent study [1] also shows that our traditional homogeneous network is rapidly evolving into a heterogeneous one. However, fairness among heterogeneous TCP variants has not been evaluated in a 10Gbps high-speed networking environment. In this demonstration, we show audience fairness among heterogeneous TCP variants by using CRON [2], a 10Gbps high-speed optical network testbed. CRON is a cyber-infrastructure with reconfigurable optical networking environment that can provide multiple networking testbeds consisting of routers, delay links, and high-end workstations operating at up to 10Gbps bandwidth. Different application developers and networking researchers can use those virtually created high speed networking and computing environments without technical knowledge of network hardware and software. With the support of NSF (award #0821741) and GENI grants, the LANET network research group at Louisiana State University has developed and operated the CRON testbed, which is based on the framework of Emulab project [3]. CRON provides integrated and automated access to a wide range of high-speed networking configurations, such as National Lambda Rail (NLR) [4], Internet2 [5], Louisiana Optical Networks Initiative (LONI) [6], as well as purely user-defined networks. Based on specific demands, users can dynamically reconfigure whole computing resources, such as operating system, middle-ware, and applications. Because of the automated and reconfigurable characteristics, all types of experiments over CRON could be repeatable and controllable. We setup dumbbell topology with three pairs of TCP senders and receivers in CRON as shown in Figure 1. The three pairs of senders and receivers have three different highspeed TCP variants, namely TCP-SACK [7], HighSpeed TCP (HSTCP) [8], and CUBIC TCP (CUBIC) [9]. We show the significant difference of throughput among these three different TCP variants in 10Gbps high-speed networking environment. Thus, audiences can have a clear view of the fairness among heterogeneous TCP variants over 10Gbps high-speed networks. Then, we change queue discipline in intermediate router to RED (Random Early Detection) [10] and CHOKe (CHOose and Keep for responsive flows, CHOose and Kill for unresponsive flows) [11]. We show that active queue management (AQM) schemes can tame the unfairness among three different TCP variants. Router buffer size is changed as well to observe the difference in fairness behavior. During the demonstration, we set the router buffer size to different Bandwidth-delay product (BDP), such as 10% of BDP and 20% of BDP. These different router buffer sizes show different fairness behavior to the audience. This demonstration will showcase the configuration of 10Gbps networking environments. Fairness among heterogeneous TCP variants in 10Gbps high-speed networks will be presented to audience. More works can also be demonstrated if time allows, such as Linux parameters tuning for 10Gbps environment, queue discipline parameter setting, network delay setup, multiple long-lived TCP flows, short-lived TCP flows generation, RTT fairness among heterogeneous TCP flows, and etc.