[DNA] Link-layer notifications in the presence of RSTP (802.1/802.3) (fwd)

[Bernard Aboba <aboba@internaut.com>]

Here are some comments by Mick Seaman of IEEE 802.1.

>From: Mick Seaman <mick_seaman@ieee.org>
>Reply-To: mick_seaman@ieee.org
>To: STDS-802-1-L@listserv.ieee.org
>Subject: [802.1] Link-layer notifications in the presence of RSTP
>(802.1/802.3)
>Date: Wed, 8 Jun 2005 16:03:43 -0700
>
>Paul Congdon notified the 802.1 email list of the existence of
>
>http://webcamserver.eng.monash.edu.au/~warchive/dna/2005-06/msg00088.html
>
>which discusses link up/down for 802.3/802.1 networks.
>
>Here are a few thoughts/information points on the subject:
>
>When a 802.3 link (LAN) comes up and at least one of the attached stations
>is a bridge (port) running RSTP, with the other being another bridge
>running
>RSTP or an end station,  then the following cases will determine the delay
>to, and knowledge of "link up" (i.e. bridge port state Forwarding, to be
>specific).
>
>A. The LAN is point-to-point (as per auto negotiation or forced by
>management - adminPointToPointMAC in a Bridge), or shared media
>
>B. The setting of the RSTP AdminEdge and AutoEdge RSTP parameters, which
>the
>RSTP state machines use to determine  operEdge i.e. am I connected to
>another bridge?
>
>Leaving shared media on one side at the moment, as being of minority
>interest currently, and just looking at point to point, scenarios include:
>
>PP1: Point to point, RSTP bridge to end station, with bridge with AdminEdge
>TRUE:
>
>     Bridge brings link up immediately.
>
>PP2: Point to point, RSTP bridge to end station, with bridge with AdminEdge
>FALSE and AutoEdge TRUE:
>
>     Bridge brings link up after 3 seconds approx.
>
>PP3: Point to point, RSTP Bridge to RSTP Bridge. AdminEdge and AutoEdge
>both
>don't care.
>
>     Both bridges will bring link up after 0-2 seconds, as the RSTP
>Proposal
>Agreement handshake will work in that time.
>
>PP4: Point to point, end station to end station.
>
>     End stations can assume link up after 3 seconds, if no better
>knowledge
>is available.
>
>The following are potentially useful end station behaviors in the short
>term, not specified in 802.1D because that doesn't specify end station
>behavior.
>
>ES1: Receive BPDUs. If none are received with 3 seconds, then assume link
>is
>up. That will deal with legacy STP Bridges. If an RSTP BPDU is received
>then
>declare link up if and only if the Forwarding flag in the BPDU is set, that
>allows bridges with AdminEdge TRUE to be autodetected by the end station.
>If
>an STP BPDU is received wait the requisite long time to bring the link up.
>
>ES2: Send RSTP BPDUs with lowest priority and the Agreement flag set, that
>should cause the/any RSTP Bridge to bring the link up - which can be
>detected by the end station as per ES1. (This is probably not the right
>long
>term behavior).
>
>ES3: Use .1AB to advertise "I am not a bridge". This would couple with
>bridge behavior.
>
>The following is potentially useful bridge behavior:
>
>BR1: Use receipt of a .1AB "I am not a bridge" message to autoset
>AdminEdge,
>or directly to nudge the RSTP state machine in the same way. Taken as a
>package ES3, BR1, and ES1 do well.
>
>A longer term view of link-up/down should take into account the probability
>of Port-based Network Access Control being used to regulate end station
>access to the network. This requires a much better focus on what is really
>meant by "link up" -- "up" from whose point of view and to do what? The
>preferred placement of .1AB and RSTP is above MAC Security. If the security
>framework returns authentication and authorization attributes, possibly as
>Radius attributes to a bridge port functioning as the network access port,
>then both an RSTP capable bridge and an end station newly attached to a
>bridge port should be able to unambiguously answer the questions "am I
>attached to a bridge/am I attached to a station that I will allow to
>operate
>as a bridge?". The bridge port can come "up" as soon as soon as it knows
>that it is attached to an end station, which should be just as .1AB and
>RSTP
>are about to run. Behavior ES1 may still be useful to mark the exact start
>of link up, though ideally the completion of the distribution of session
>keys would mark "link up" except in those cases where further active human
>intervention would be required.
>
>Mick
>
>IEEE 802.1 list info: http://www.ieee802.org/1/email-pages/hngih305.html