> there is a fundamental tradeoff between latency and throughput
Only for the unpredictable part of the load. If you can predict it, you can have both. For that you need to control the entire stack though, starting from the OS and the apps. That's what the "scalable congestion control on the sending host" is about.
Only for the unpredictable part of the load. If you can predict it, you can have both. For that you need to control the entire stack though, starting from the OS and the apps. That's what the "scalable congestion control on the sending host" is about.