itojun@iijlab.net writes:

> >It is my understanding that Linux now supports multiple routes to a
> >destination (often referred to as multipath routing).  Multipath routing 
> >is more powerful, but that doesn't make BSD less ``correct'', just less
> >powerful.
> >
> >The reason Linux does it and BSD doesn't is because people are actively
> >overhauling the linux routing code and no-one seems to have stepped up
> >to do the same with NetBSD (*BSD whatever..).  There are many people
> >(including myself) that would love to do this but don't have the time.
> 
> 	I agree it would be nice, but we need to be very careful designing it.
> 
> 	If I understand correctly, multiple destination support needs a
> 	careful consideration.  We cannot just forward packets at random to
> 	destinations (you may shuffle packets due to different path and
> 	and it makes reassembly code less performant).  And it is not really
> 	clear enough what is the best way for kernel to pick a path from
> 	equal-cost multipaths.

Your right. Here is a draft I co-authored on the subject of ECMP:

http://search.ietf.org/internet-drafts/draft-thaler-multipath-04.txt

I'll be revising the Hash-Threshold parts (and the referenced draft) in
the next round which will hopefully finish both of them off.

Chris.