[rbridge] Time to summarize "forward or block" BPDU thread
saikat at seas.upenn.edu
Sun Oct 16 17:09:44 PDT 2005
It is probably a standard scenario in IS-IS protocol (and consequently a
standard solution is known), but I was thinking that if the RBridges
disregards BPDUs entirely, then in some situations loops may form for some
period. For example, suppose $T_1$ and $T_2$ are disjoint trees formed
entirely by legacy elements. $R_1$ and $R_2$ are the designated RBridges for
the "links" $T_1$ and $T_2$, respectively. At this point, there is no loop
in the system. Now suppose that a physical link (or a legacy bridge) is
added that connects a legacy bridge of $T_1$ to a legacy bridge of $T_2$.
This addition will trigger a new Spanning tree computation, which will
result into a single spanning tree that spans the nodes of $T_1$ and $T_2$.
At this point, effectively the RBridges $R_1$ and $R_2$ are connected by a
single "link" and until one of the RBridge gets "de-designated", there would
exist a loop. Two questions arise in my mind: (i) how fast would that
"de-designation" process be? And (ii) would it be useful (i.e. would it
accelerate the convergence) if RBridges "listen" to the BPDUs?
From: rbridge-bounces at postel.org [mailto:rbridge-bounces at postel.org] On
Behalf Of Radia Perlman
Sent: Sunday, October 16, 2005 7:12 PM
To: Developing a hybrid router/bridge.
Subject: [rbridge] Time to summarize "forward or block" BPDU thread
It looks like this thread was discussed with two different subject
lines, and it's also
gone on long enough it's time for someone to summarize it.
I strongly believe that things will be more stable if RBridges do NOT
we decouple the protocols.
That TRILL is kind of like a layer between bridging and routing.
If RBridges do not forward BPDUs, then whether or not the TRILL link
state protocol is
converged, it won't affect the little spanning tree inside a particular
That way, the little spanning trees will converge as quickly as possible
(because it's small),
and cannot possibly be disrupted by RBridges wormholing BPDUs.
I do not understand the reasons why people want to forward BPDUs. I
think the arguments (which
I may not be presenting fairly because I don't understand them) are:
a) you'll get a more optimal combined spanning tree on which
unknown/broadcast frames will
be sent if it is one global spanning tree, rather than little
independent spanning trees hooked together
by the independently computed spanning tree calculated by IS-IS.
b) forwarding BPDUs, and having a global spanning tree, will prevent loops.
I strongly disagree with both those arguments. For a) What do people
think is more optimal about a tree
computed that way vs having independent trees inside each bridged
island? And besides, there isn't
a single spanning tree...it's a spanning tree per ingress RBridge.
For b) no...I believe that having both the RBridge algorithm and the
bridge algorithm feeding into
each other will create much slower convergence.
If I haven't summarized correctly, then someone else can try to capture
the arguments, but I do
think we should merge discussion under one subject line, and restart
with a summary.
rbridge mailing list
rbridge at postel.org
More information about the rbridge