Sharing an interesting paper I bumped into..
> A fork() in the road

> https://www.microsoft.com/en-us/research/uploads/prod/2019/04/fork-hotos19.pdf  

in particular Figure 1 - fork versus spawn performance

And something I didn't realise... 

> a process using, DPDK with a kernel-bypass NIC [12], or OpenCL with a GPU, 

> cannot safely fork since the OS cannot duplicate the process state on the NIC/GPU.

> This appears to have been a continuing source of bafflement 

> among GPU programmers for a decade at least

cheers -ben