> The basic idea would be to allow each non-global zone to have its own > name space for network devices. For example, this would allow each zone > to have a "net0" interface. Each "net0" could map underneath to separate > physical devices, VNICs, etc. As Erik mentioned, this means that > obtaining the underlying device would have to include the zone id in > addition to the device name. > > This would make zone configuration and eventual migrations easier since > the configuration refering to network interface names could be identical > for all the zones on a machine, and would ease the migration since we > wouldn't have to worry about conflicts with other zones running on the > target host.
I agree this is compelling, and very much philosophically in-line with the Clearview vanity naming model. However, I am not sure how to make it administratively "approachable", unless we're comfortable saying that links in local zones are not visible from the global zone. My recollection is that restriction *would* be consistent with the administrative model for stack instances (e.g., that one must log into the zone to administer its networking stack) -- if so, this might not be conceptually problematic. However, it does seem a bit at-odds with things like ps(1), which allows processes from all zones to be manipulated. But that approach is also not without its problems (e.g., if different zones are running different name services, a ps(1) in the global zone may return some misleading results). Perhaps we should touch base with the Zones team. -- meem
