Investigating high-speed transport protocol flows on 10-gigabit testbed JGNII; with monitoring the Linux kernel internally

The recent deployment of high-speed network equipment has made it possible for us to enjoy the benefit of high-speed transport protocols in 10-gigabit-class networks. However, a TCP flow in such high-speed environments might cause extremely bursty traffic that could create a bottleneck anywhere in the end-to-end path, i.e., sender/receiver hosts could be bottlenecks. Therefore, we have been developing TCPVisible, an easy-to-use tool for monitoring TCP/UDP-related internal dynamics at sender and receiver hosts in detail, which allows us to know when and where packet losses have occurred internally in the kernel. We introduce preliminary experiments in this paper investigating the performance of two kinds of high-speed transport protocols (HTCP and BIC) in 10-Gbps paths between the US and Japan, where the usefulness of TCPVisible could be demonstrated.