RSTP IEEE 802.1w

[Samsoum]

How many switchs can a RSTP IEEE 802.1w can have in total? and any documentation ?

Thanks

[M02@rt37 · Accepted solution]

Hello @Samsoum 

40 switches according to https://www.etherwan.com/sites/default/files/setting_up_stp-rstp-mstp_and_alpha_ring.pdf

but

from my point of view, this guideline may not necessarily reflect the IEEE 802.1w standard...and if we refer to that standard, there isn’t a fixed maximum number of switches defined for an RSTP network. Instead, the standard provides a guideline regarding the network diameter—that is, the maximum number of bridge (mean switch) hops between any two network nodes. The standard recommends that the network diameter should not exceed 7 hops. 

[Joseph W. Doherty]

The actual max number of switches with STP is a function of the topology, and the capability of the switches and STP parameters.

For example, in classic STP, 8 switches in a ring topology would be problematic but a star topology, using something like the old 6513s with 96 port line cards, could have over a thousand switches without extending the star topology with more than 2 layers!  (For the latter, you likely run into other practical issues way before hitting even a default STP logical limit - as would be the case for 8 switches in a ring.)

The diameter 7 limitation, mentioned by M02@rt37 , applies to classic STP with the usual defaults.  Rapid-STP, in Cisco's implementation, has a default hop limit of 20 which can be configured up to 40.  Rapid-STP, should, in theory, better support a larger diameter topology, but both have practical scalability issues.  There are other L2 protocols, like REP, more suitable for other cases, like large ring topologies.

[Ramblin Tech]

IEEE 802.1D-2004 (802.1w RSTP was incorporated into .1D with 2004 edition) allows a maximum time of 40 seconds for BPDU messages to propagate from the root to the farthest switch (refer to maxAge value in Table 17-1). The standard also provides for an absolute maximum of 1.0 sec for a switch to receive & process a BPDU before forwarding out all appropriate interfaces. By my understanding, if the root is at the “edge” of the network, then the max diameter would be 40 hops (@ 1s/hop); if the root is at the center, then the radius could be 40 hops.

What does that say about the max number of switches? Nothing, really, since a root at the center of the network could conceivably have 10, 100, 1000 interfaces or more, each spreading out in with a radius of 40 hops. And what if each hop only spent 0.1s processing each BPDU instead of 1.0s, could you then have 10x the number of switches? What if it spent 0.01s?

Bottom line: the size of a STP network is not limited by the standard, but by good judgement, which says keep it as small as possible. Large, flat L2 networks were initially popular in the ‘90s, but soon fell out of favor as they kept melting down. A certain San Jose based router company achieved great success by preaching “route where you can, bridge where you must”.

[Joseph W. Doherty]

"What does that say about the max number of switches? Nothing, really,"

Exactly, regarding a logical limit.  I.e. it's not 40 in total.

"Bottom line: the size of a STP network is not limited by the standard, but by good judgement, which says keep it as small as possible."

There are actual limits imposed by the standard, but some, like actual timing considerations, can be hard to correctly predict.  Consider the impact of congested links and/or a very busy control planes.  Such lead to unexpected meltdowns.  Consider my prior example of 1,000+ switches, in just two layers, with 1,000,000 edge ports, without portfast configured.  Logically, no issue, but physically?

So, good judgement depends on good knowledge and a desire to avoid discovering the proverbial straw that breaks the camel's back.