Those are real-world speed bumps that make sense when you want to do stateful processing w/ large working sets, but not really for this "just get packets throug

Those are real-world speed bumps that make sense when you want to do stateful processing w/ large working sets, but not really for this "just get packets through the stack" toy benchmark that fits in cache and the workload is parallel-friendly. And also the article is talking about handling just 10G, not 100G!