> * The result of this optimization suggests that there is no way to make 
> generation of 20-bit and larger numbers efficient enough to do in the packet 
> forwarding path, if a modular exponentiation is involved.  However, we might 
> be able to reach a compromise where some bits are pure LCG output (or e.g. 
> the output of one LCG stuff into a power-of-2 LCG) and the rest are generated 
> by exponentiation.

	fragment ID won't be assigned on packet forwarding path.  it is
	generated only in ip_output().  so the above "no way to make..."
	does not make sense (no need to worry).

itojun