Re: refactoring ip_output() and the L2 _output()

[Joerg Sonnenberger <joerg%britannica.bec.de@localhost>]

On Mon, Feb 04, 2013 at 10:10:01AM -0600, David Young wrote:
> On Mon, Feb 04, 2013 at 02:57:57PM +0100, Joerg Sonnenberger wrote:
> > On Mon, Feb 04, 2013 at 08:26:47AM -0500, Greg Troxel wrote:
> > > I don't think it would be an improvement - you'd need a separate table,
> > > and you'd need to do two lookups, one to find the cloning route, and
> > > then to find the arp entry.  Right now the arp entry, if present, is
> > > found directly.
> > 
> > I disagree. You are ignoring the impact of having the leaf nodes in the
> > routing table, which can be quite significant.
> 
> Please explain more about this.  It's not clear what assumptions about
> the routing data structure you have made.

Moving the ARP entries out of the main routing table makes the PATRICA
tree more shallow. My assumption here is that a plain hash table is no
worse in performance than adding at least one level of the PATRICA tree,
when collisions are not an issue.

Joerg