on the network throughput front, we’re fighting (albeit in distributed manner) for getting throughput from commodity PC hardware on par with dedicated, hardware routing platforms. with OSes like Linux and BSD.
to that end, recent document published after last Linux Congress in Hamburg shows that while it’s important to select proper multi-core CPU and motherboard to do fast traffic forwarding, we’re still hitting bottleneck at around 1Mpps. curiously enough, on one of the slides you can spot information, that large FIB in Linux doesn’t impact performance too much. you can’t say that about some of the older, legacy solutions using different FIB tricks - like for example compression. you can also find some interesting information about Linux kernel tuning. you may remember than years ago NetFPGA was published, using Xilinx chipset. it’s still 4 GE ports, but quite efficient and with readily available SDK.
...