Is copying network packets really the biggest overhead in perlbal? I would have thought more likely it's just perl data structures getting copied all the time.

Though I guess any small win is worth it with the amount of data you're copying around.

I know where Perl uses CPU, so I'm pretty careful to avoid painful stuff. I believe the reads/writes syscalls were actually a big part of the last profile I did.

Good stuff. You've only been talking about wanting that for at least three years now (as long as I can remember).

How long before that hits mainstream (Debian/Ubuntu/RHEL/CentOS/FC/whatever) kernels that are part of main distributions?