080319 Notes

3/19 10:30 NWAM design summit finish
2 - IPv6 NCUs should always be enabled
13 - can we torch network-service?
        maybe short circuit net-svc in first case when nwam enabled?
        ifconfig -AUd4 is hack at best anyways (netmasks, dns)
                so deal with it in NWAM
        but need to make ipqos code its own service - talk to ML
        phase 0 bug - defer and note that you can restart NWAM
        could we do that by making net/service <- net/physical:default
                but that fails due to dependencies on net/service
                so can we create net/service:{default,nwam}
                does that poliferate?
14 - iptun <- dependent on network/physical:default since it is old world
        -- need to talk to seb
3 - have automatic vs user ncp, bimodal switch
        can it be merely default behavior - alan (agree mph)
        renee - maybe node ncp but mode
        no ncp => automatic mode
        if somebody sets variable and there isn't any config then the
                template config is written out
        the smf variable always tells us what to do
        when we install ncp then we set the variable to true but the
                user can reset to false which then says ignore ncp
        this gets complex in switching back and forth so instead go into
                maintenace mode
                can we create customized error codes?
                if user create ncp then turn off but then can be controlled
                        independently
1 - activation
        since we have knowledge of over then you can't point at vanity
                name and then later point vanity name at another link
        by skipping network/physical:default are we missing vanity names
        two rename cases
                - changing label
                - applying old name (dr'd) to new interface
        cfgadm/rcm has seperate policy management parts
        rename - can't rename to current thing or from remove thing
                (dead bodies are frozen)
        dr framework more tied to boot infrastructure (hostname files, etc.)
        future - remove active part of dr in future, but now need a way to
                get dr out of the way when nwam is active
        how does NWAM restore its view of the work wrt link rename
                - problably linkid
        nwam and dr can't cross paths?  too strong?
                in any case need way to have rcm not act on events
        sysevents for new vlans etc is caused by rcm so if we do this
                then who is responsible for causing those sysevents
                common code in libdladm
        email to david about what he has done in this area

1G