Update: The built-in irq balancer is poor. It puts all irq load from both active nics onto a single thread. Once I manually tuned SMP affinity to manually assign each NIC IRQ to a different core, I got an instant 50% improvement in throughput (to 750 Mbits).
http://www.alexonlinux.com/smp-affinity-and-proper-interrupt-handling-in-linux
4 Likes