On Fri, Feb 09, 2007 at 09:14:27PM +0100, Tonnerre LOMBARD wrote: > Salut, > > On Mon, Jan 29, 2007 at 10:45:08AM +0100, Claudio Jeker wrote: > > Note: the OpenBSD routing table does not do that. > > It's hard to do hardware accelerated FIBs without the hardware, isn't it? >
Using a compiled FIB may be even useful in software. e.g. an LC trie needs around 3-5MB for a full view instead of the 25+ MB of the patricia trie. The smaller size results in less CPU cache trashing and higher speed. Btw. Cisco CEF is nothing more than a compiled FIB everything is still done in software. > > While IPv6 has a static header size it uses header stacking and so every > > router has to do the same stupid header parsing that needs tons of special > > logic. > > If you need to look at them at all, that is. For simple end-to-end routing, > this is not required. > If you don't look at the additional IPv6 headers then you should do the same for IPv4 and we're back on square 1. It is in the standard and needs to be implemented even if only 1ppm of the transported packets are using it. -- :wq Claudio
