[c-nsp] Limits of STP/RSTP/REP?

[Ross Vandegrift]

On Wed, May 20, 2009 at 08:09:17PM +0200, Garry wrote:
> Wondering, what's the sensible limits of STP, RSTP, REP or any other
> spanning tree/ring protocol available on Cisco switches like 29, 35, 37
> or ME3400 series? I was told by a customer whom we try to sell some
> Cisco gear that beyond anything like 4 or 5 switches in a ring,
> recognition/recovery times of the ring would quickly go well beyond 10s
> on failure of a link ...
> 
> Now, on STP the times are definitely somewhere in that range, but what
> about RSTP or stuff like REP?

In a usual ring scenario, an RSTP bridge would have two paths to root.
For either cost or tiebreaker reasons, one would be chosen as the root
port and one would be chose as the alternate.  If the root port goes
down then the switch will rapidly move the alternate port to root and
start sending BPDUs with the TCN flag set.  This will in turn cause
the other bridges to age-out their MAC table.

The process of flooding TCNs should lead to worst-case full
reconvergence in approximately (hello-interval * max number of hops in
the active topology) seconds.  For a ring of five switches, all of who
have only edge ports on the non-ring interfaces, this is a worst case
of six seconds until all bridges have flushed their MAC tables.

This worst case isn't very realistic.  802.1D-2004 specifies that a
bridge should immediately transmit a BPDU if it has new information,
including topololgy changes.  It also specifies that there is an
absolute max of 1.0 seconds permitted between an external event
requiring a BPDU and the transmission of that BPDU.  That gives a more
realistic worst-case of three seconds. [1]

In practice, I've seen tons of link failures happen with no measurable
impact on traffic.  That doesn't mean there wasn't any, but it's quick
enough that I don't usually drop a ping during link maintenance.  Of
course this isn't a super-fine measuring tool, so do your own testing!

Cisco has a decent article on various features and changes in RSTP
that you might find interesting:
http://www.cisco.com/en/US/tech/tk389/tk621/technologies_white_paper09186a0080094cfa.shtml

And finally, if that isn't enough, the IEEE lets you download 802
specs for free at http://standards.ieee.org/getieee802/.  802.1D-2004
(and .1Q too) is actually pretty readable for someone serious about
understanding it.

[1] Of course, it's anyone guess as to how close the vendors come to
implementing this :).


-- 
Ross Vandegrift
ross at kallisti.us

"If the fight gets hot, the songs get hotter.  If the going gets tough,
the songs get tougher."
	--Woody Guthrie