> >   However, there will be cases where a given link property is either not
 > >   generally applicable, or where we need to gain more experience before
 > >   committing to an administrative model.  For these cases, we propose to
 > >   introduce a special namespace, starting with "link_"[1] that is reserved
 > >   for properties that are specific to a given link and subject to change
 > >   at any time.  Accordingly, no guarantee is made to preserve behavior
 > >   across an upgrade.  This distinction will be documented in dladm(1M) and
 > >   any link-specific tunables will be covered in the appropriate network
 > >   driver manpage.
 > > 
 > >     [1] The actual driver name should be omitted -- e.g., "link_xyzzy",
 > >         *not* "link_bge_xyzzy".
 > 
 > This actually seems to define the Brussels "Private Property".
 >
 > > This was originally added at the request of the Neptune development team,
 > > but I don't think they are using it yet.  I'm OK with doing something
 > > different, but there needs to be some way to (a) indicate interface
 > > stability and (b) provide link-specific tunables.
 > 
 > I recall having this discussion before (see 
 > http://mail.opensolaris.org/pipermail/brussels-dev/2007-June/000186.html).
 > Seems like the conclusion that was reached was that private properties
 > should be documented in the driver's 7D man page, along with the 
 > associated interface stability.

So what keeps the driver's private property names from colliding
non-private names that are later added?  (That's one of the intents of the
the link_ prefix.  The other is making the division between private and
public property names an explicit part of the administrative experience,
so the admin's expectations remain appropriately set.)

-- 
meem

Reply via email to