On Mon, 2009-03-23 at 14:46 -0400, Sowmini.Varadhan at Sun.COM wrote: > On (03/23/09 14:31), Sebastien Roy wrote: > > > Also for example the 'ip_fragment_timeout' which needs different timeout > > > value for v4's and v6's interfaces can be easily achieved using the '-f' > > > flag instead of defining two different variables. > > > > I don't see the problem with different variables. The variables are > > specific to a protocol operating over an IP interface. The model that > > You are suggesting that the interface name-space have the protocol > embedded in it (which was the original Mentat/ndd approach).
I'm suggesting the exact opposite... Your proposal requires that there be separate IP interface namespaces for IPv4 and IPv6 (as in the current ifconfig-based administrative model), and I'm asserting that this model is busted. There should not be separate IP interface namespaces for IPv4 and IPv6. > This > also introduces artificial constraints. Taking the classic frag_timeout > example, it's more cumbersoume to have to specify > "set-prop ip_frag_timeout <foo>" and "set-prop ip6_frag_timeout <foo>", > when I really want all IP packets (V4 and V6) to have a frag timeout > of <foo>. Users aren't constantly setting link properties 24 hours a day 7 days a week. They're setting a property or two once and that setting will sit there virtually forever. In any case, given that there is a single IP interface namespace (and I think that we agree that this is what we want, right?), then I'm the one who is confused about the property namespace. Can this be explained in more detail? -Seb
