[Rstplib-users] RE: 802.1w: double flushing in simplest case ?

[Mick S. <mic...@ie...>]

I agree with Shyam that correctness is the goal here. Efficiency in flushing
is essentially an implementation issue (of a single bridge) and the standard
is not responsible for dumb implementations where every entry has to be
physically removed from a database to flush it (and the databse
reorganized).

At the absolute minimum each entry in a well designed database should have a
DELETED bit, because it is almost a certainty that it will be relearnt
unless the station or bridge has been removed from the net. It is perfectly
possible to design a database such that it can be flushed by changing a
single central variable (just need a couple more bits on each entry).

This really an implementation discussion since the number of options is so
wide, especially with 'flow' based databases that extend the bridges
functionality. Some tradeoff aganist the speed of relearning is also
possible if you are realy stuck. Obviously if you expect two messages
carrying a topology change a little bit more flooding would allow you to
wait for the second before starting learning again. A wide variety of
options are there for anyone who cares to do the performance analysis. This
is of course best done before building your bridge, but a bridge that was
efficient at database operations to support earlier .1D should continue to
be so.

Mick

-----Original Message-----
From: own...@ma...
[mailto:own...@ma...]On Behalf Of Shyam Kaluve
Sent: Sunday, January 06, 2002 11:10 PM
To: Ar...@op...
Cc: std...@ie...; rst...@li...
Subject: RE: 802.1w: double flushing in simplest case ?



Alex,

When a p2p link comes up and becomes part of the active topology, both
ports connected by that link become topology change initiators. The set of
ports that will be flushed for each of the initiators will be different,
but yes, there will be a substantial number of ports common to both the
sets and flushed twice.

For correctness you will need to flush the minimum set of ports only once.
But since this operation is performed by each bridge in its own pace, once
may be not enough. The residual traffic as the topology is converging may
cause some ports to learn wrong destinations. So IMHO the standard should
lean towards correctness than efficiency in this case. Next generation of
bridge hardware will be more 802.1w friendly so this question may become
moot.

-shyam

Note: All disclaimers apply.

At 10:24 AM 1/6/2002 +0200, Alex Ruzin wrote:
>Ladies & Gentlemen !
>Excuse me, routers & brouters are very interesting devices,
>but what about my original question (double flushing in regular bridges) ?
>Best regards, Alex