[rbridge] Ingress Rbridge address and BCN - 2nd Issue
caitlinb at broadcom.com
Sat Oct 28 18:07:00 PDT 2006
Gray, Eric wrote:
> Have we moved into the realm of _requiring_ support for
> BCN in all RBridges?
> We can assume the "core RBridge" is BCN capable if it
> has triggered a BCN event, but we cannot assume that the
> ingress is likewise BCN capable unless we're going to require
> BCN support in every RBridge.
Consider the following scenario: Host A is 802.1au compliant
(it will respond to BCNs), and 802.1au capable Bridge C is
part of its path to the end destination D.
At this time if C generates a BCN to A, A will receive it
and adjust its sending rate in compliance with 802.1au,
without dropping frames.
If we place an 802.1au-ignorant RBridge B is injected
between A and C then congestion domain has to shrink,
and C has to act as an edge to the congestion domain.
C has no method of regulating the flow from A to D
other than dropping frames.
If RBridge B is nominally 802.1au compliant, but will
not act as a proxy to relay BCNs then it effectively
becomes the edge of the congestion domain. That means
that B must regulate the flow from A to D by dropping
The only way an RBridge can avoid impairing 802.1au
congestion avoidance is for it to fully proxy 802.1au
functionality when interacting with an 802.1au bridge
that is not an rbridge.
More information about the rbridge