Re: [DNA] Definition of "Link Up" and "Link Down" events?

[Bernard Aboba <aboba@internaut.com>]

> But is this really of concern to IP? The IP layer just wants to know whether
> or not the link layer thinks there's a high probability of getting a packet
> across. The details of how that's done are indeed part of the state machine
> in the driver, but are they of any concern to the IP stack?

"High probability" is different from "the link layer is in a state where
IP packet transmission is not possible".  The former depends on link
quality, whereas the latter depends on the link layer state machine.  As a
result, the former is likely to be highly transient, while the latter is
likely to be less so.

> Right, that's what I was trying to get at with the statement on policy. In
> the case you've cited, the NIC policy is that extensive frame loss is not
> sufficient to trigger a "Link Down". That might not always be the case, for
> example, other NIC cards might have different policies, or the policy might
> be configurable.

Experiments do indicate that NICs behave quite differently, and some of
that difference is due to different approaches taken by reasonable
implementations as opposed to obviously broken behavior.   For example,
better 802.11 NIC implementations start scanning after only a few
retransmissions (~3) since 3 consecutive lost frames is unlikely to be
caused by collisions.  These implementations still obey the 802.11
default shortretrylimit (7) so they continue retransmitting
even as the NIC is interleaving scans to find another potential point
of attachment.  In most cases a "Link Down" is not signalled unless
frame loss continues and another suitable AP cannot be found within the
ESS.

> >Sure, but I'd suggest that we should be clear that this additional info is
> >not a determinant of "Link Up" and "Link Down" events.
>
> Hmm, what about the case where 802.1x authentication must complete prior to
> "Link Up"? That has nothing whatsoever to do with the quality of the link
> from a frame transmission standpoint?

Since EAP supports retransmission, and on 802.11 802.1X frames are sent to
a unicast address, it is possible for 802.1X authentication to complete
even in situations where there is high frame loss.  So it is possible for
an 802.11i link to reach the "Link Up" state even though link quality is
poor.