TCP Rapid: From theory to practice

摘要: Delay and rate-based alternatives to TCP congestion-control have been around for nearly three decades seen a recent surge in interest. However, such designs faced significant resistance being deployed on wide-scale across the Internet — this has mostly due serious concerns about noise delay measurements, pacing inter-packet gaps, and/or required changes standard stack/headers. With advent of high-speed networking, some these become even more significant. In paper, we consider Rapid, proposal ultra-high speed congestion control, which perhaps stretches each challenges greatest extent. Rapid adopts framework continuous fine-scale bandwidth probing, requires potentially different finely-controlled gap every packet, high-precision timestamping received packets, reliance gaps. While simulation-based evaluations show that it outstanding performance gains along several important dimensions, will not translate real-world unless above are addressed. We design Linux implementation after carefully considering challenges. Our 10Gbps testbed confirm can indeed achieve claimed gains, would possible was