Skip to content
[menu][close]

FOUNDRYNETNo. 001SEPTEMBER 2026SWITCHING

MSTP: Regions, Instances and the Multiple Spanning Tree Protocol

MSTP lets a switched network use both uplinks by running a few spanning trees instead of one or hundreds. It only works when every switch agrees on the region, byte for byte.

On this page

MSTP (Multiple Spanning Tree Protocol) runs a small number of spanning tree instances and maps many VLANs onto each one, so a switched network can block a different set of links for different groups of VLANs without running a separate tree per VLAN. Each instance uses the rapid convergence of RSTP, and every switch that agrees on the same VLAN-to-instance map forms one MST region that behaves like a single bridge to the outside.

MSTP was published as IEEE 802.1s-2002 and folded into IEEE 802.1Q in 2005; the current text lives in IEEE 802.1Q-2022. The basics of root election, port roles and RSTP handshakes are on the spanning tree protocol overview. This page is about what MSTP adds: regions, instances, the configuration that must match on every switch, and how it fails.

01 Why run MSTP instead of RSTP or per-VLAN spanning tree?

Plain 802.1D or 802.1w builds one tree for the whole switched network. With two uplinks from an access switch to a distribution pair, one uplink carries every VLAN and the other sits blocked until the first fails. Half the uplink capacity in the building does nothing.

Per-VLAN variants such as PVST+ and Rapid PVST+ fix that by running one full spanning tree per VLAN, so you can make distribution switch A the root for odd VLANs and B the root for even VLANs. The cost grows with the VLAN count: 300 VLANs means 300 instances, 300 sets of BPDUs on every trunk and 300 state machines on every switch CPU. They are also proprietary, and they meet standards-based switches only through a single common tree.

MSTP sits between the two. Most designs need only two or three distinct forwarding patterns, so MSTP lets you build exactly that many instances and assign every VLAN to one of them. Two instances can split the load across both uplinks while the switch runs two trees, whatever the VLAN count.

Spanning tree variants compared
VariantStandardTreesConvergenceUses redundant uplinks
STPIEEE 802.1DOne for all VLANs30 to 50 s, timer basedNo
RSTPIEEE 802.1w, now 802.1D-2004One for all VLANsAbout a second, handshake basedNo
PVST+ / Rapid PVST+ProprietaryOne per VLANTimer based / about a secondYes, per VLAN
MSTPIEEE 802.1s, now 802.1QOne per instance, VLANs mapped inAbout a second (RSTP per instance)Yes, per instance

02 What is an MST region?

An MST region is a set of connected switches that agree on three values, and the agreement has to be exact:

  • the region name, a text string of up to 32 characters;
  • the revision level, a 16-bit number;
  • the VLAN-to-instance mapping, which the switch does not send in full. It sends a 16-byte configuration digest computed over the mapping table for all 4096 VLAN IDs.

Every BPDU carries all three. A neighbour whose name, revision or digest differs is in another region, even if the difference is one VLAN mapped to a different instance or a name that differs only in letter case. The port facing it becomes a region boundary port, and only the common tree crosses it.

Inside a region there are two kinds of tree:

  • The IST (internal spanning tree), instance 0. It carries every VLAN not mapped elsewhere, and it is the region's part of the CIST (common and internal spanning tree) that ties regions, plain RSTP switches and legacy STP switches into one loop-free topology. The Junos MSTP documentation describes the CIST as treating each region as one virtual bridge, regardless of how many switches are inside.
  • MSTIs (multiple spanning tree instances), numbered from 1. Each has its own root, its own priorities and port costs, and its own port roles. The BPDU format allows up to 64 MSTIs per region; individual platforms may support fewer, so check the data sheet before you design for many.

Because each instance computes its own tree, one physical port can hold different roles at once. An access switch uplink to distribution A can be the root port in instance 1, where A is root, and an alternate port in instance 2, where B is root, forwarding frames for VLANs 10 to 1999 while discarding frames for VLANs 2000 to 3999. Port state is per instance, not per port, which is exactly what lets both uplinks carry traffic. A few properties stay per port across every instance: whether it is an edge port, whether the link is point-to-point, and whether BPDU guard or root guard applies. A topology change in one instance flushes MAC addresses only for the VLANs mapped to that instance, so a flapping link in instance 2 does not disturb forwarding tables for instance 1.

MSTI information never leaves the region. One BPDU per port per hello carries the CIST data plus a record for each MSTI, so adding an instance adds a few bytes to an existing BPDU rather than a new stream of BPDUs. MSTP also replaces the message age limit inside a region with a hop count: each switch decrements it, and the 802.1Q default maximum is 20 hops.

03 How do you plan and configure MSTP?

The work is mostly planning. The commands are short; a mapping that is wrong or inconsistent is what causes outages.

  1. List the forwarding patterns you need. A typical campus with a distribution pair needs two: instance 1 rooted on distribution A, instance 2 rooted on distribution B. Leave the IST for management and unused VLANs.
  2. Write the mapping for the whole VLAN range, including VLANs that do not exist yet. Map blocks such as 10 to 1999 and 2000 to 3999, so creating a VLAN later does not change the digest and split the region.
  3. Choose a region name and start the revision at 1. Use the same name everywhere; increment the revision whenever the mapping changes.
  4. Configure the region on every switch before enabling MSTP mode. Switches with the mode on and the region unset will each form a region of one.
  5. Set root priorities per instance. Make one switch the root for the IST and instance 1 with priority 4096 and its partner the root for instance 2. Give each the next value, 8192, for the other instance as backup.
  6. Align first-hop redundancy with the roots. If the distribution pair also runs a gateway protocol such as HSRP active and standby routers, make the active gateway for each VLAN the root of the instance that carries it.
  7. Mark host ports as edge ports and add BPDU guard, as with any rapid spanning tree.

On Cisco IOS the configuration looks like this:

spanning-tree mode mst
!
spanning-tree mst configuration
 name CAMPUS-EXAMPLE
 revision 1
 instance 1 vlan 10-1999
 instance 2 vlan 2000-3999
!
spanning-tree mst 0-1 priority 4096
spanning-tree mst 2 priority 8192

The partner switch gets the same region block, with priority 4096 for instance 2 and 8192 for instances 0 and 1. On Junos the equivalent lives under protocols mstp:

set protocols mstp configuration-name CAMPUS-EXAMPLE
set protocols mstp revision-level 1
set protocols mstp msti 1 vlan 10-1999
set protocols mstp msti 1 bridge-priority 4k
set protocols mstp msti 2 vlan 2000-3999
set protocols mstp msti 2 bridge-priority 8k

Uplinks bundled with LACP link aggregation count as one spanning tree port, which is usually what you want: the bundle forwards or blocks as a unit per instance.

04 How do you check that MSTP is working?

Check the region first, because everything else depends on it. On Cisco IOS, show spanning-tree mst configuration digest prints the name, revision and digest; run it on every switch and compare. The digest must be identical byte for byte. On Junos, show spanning-tree mstp configuration shows the same values.

Next, look for boundary ports. show spanning-tree mst on IOS marks ports that face another region or a non-MSTP switch with a Bound type. Inside a single-region campus there should be none. A boundary port between two of your own switches means their region settings differ.

Then confirm each instance's root. show spanning-tree mst 1 and show spanning-tree mst 2 should name different root bridges, and the access switch should show one uplink forwarding in instance 1 and the other forwarding in instance 2. If both instances block the same uplink, the priorities did not take, or the mapping on the access switch puts both VLAN blocks in the same instance.

Finally, test a failure: shut the uplink forwarding for instance 1 and confirm a ping across a VLAN in that instance loses at most a second or so of traffic while the alternate port takes over.

05 Common MSTP mistakes

  • One switch with a different mapping. A VLAN added to the map on one switch and not the others splits the region. The switches still converge, but through the CIST at the boundary, and the load split you designed disappears.
  • Changing the mapping on a live network. Every switch you edit leaves the region until the rest match, and instances reconverge. Stage the change in a maintenance window, or pre-map unused VLAN ranges so you never need to edit it.
  • Trunk allowed-VLAN lists that disagree with the instance topology. MSTP decides forwarding per instance, not per VLAN. If a VLAN is pruned from the one link its instance is forwarding on, that VLAN is cut off while the rest of the instance works. Carry every VLAN of an instance on every trunk in that instance, or move the VLAN to another instance. The VLAN design side is on the VLANs and trunk ports page.
  • Default priorities. With every switch at 32768 the lowest MAC address becomes root of every instance, often an access switch. Set the root and backup root for the IST and every MSTI.
  • Meeting PVST+ at the edge of the region. A per-VLAN switch connected to an MST region sees only the CIST. The safe design keeps the CIST root inside the MST region. Some platforms block a boundary port as inconsistent when the per-VLAN roots on the other side disagree with the CIST, which looks like an unexplained blocked uplink.
  • Too many instances. Each extra MSTI adds roots to configure and states to reason about. Two or three instances cover nearly every campus design.

06 MSTP on Foundry switches

IronWare on FastIron switches and BigIron chassis added 802.1s once it was standardised, alongside 802.1D per VLAN and 802.1w. Many Foundry networks of that era did not use it for the core, though. Rings and dual-homed aggregation pairs ran Foundry's own Metro Ring Protocol or VSRP, which converged faster than spanning tree on those shapes, and MSTP or RSTP covered the remaining access layer. When those networks are migrated, the usual target is a single MST region with two instances, and the old ring and VSRP domains become either routed links or LACP bundles. The other switching topics, from VLAN design to port authentication, are collected in the switching section index.

07 Questions

What is the difference between MSTP and RSTP?

RSTP builds one spanning tree for every VLAN, so redundant uplinks stay blocked. MSTP runs several instances, each using RSTP's fast handshake, and maps groups of VLANs to each instance, so different VLAN groups can forward over different links.

What is the difference between MSTP and PVST+?

PVST+ runs one full tree per VLAN and is proprietary, so CPU and BPDU load grow with the VLAN count. MSTP runs only as many instances as you define and is an IEEE standard, so it scales to thousands of VLANs and works across vendors.

Why are my switches in different MST regions?

Their region name, revision level or VLAN-to-instance mapping differs. Even one VLAN mapped differently changes the configuration digest. Compare the digest output on each switch and correct the one that differs.

How many MST instances should I use?

Usually two or three. Most designs need one instance per forwarding pattern, such as one rooted on each distribution switch. The BPDU format allows 64 MSTIs per region, but every extra instance adds roots to set and states to troubleshoot.

What is the IST in MSTP?

The internal spanning tree, instance 0. It carries every VLAN not mapped to another instance and forms the region's part of the common and internal spanning tree, which connects the region to other regions and to RSTP or STP switches.

Does adding a VLAN break the MST region?

Only if you change the mapping. Creating a VLAN that is already mapped to an instance leaves the digest unchanged. That is why designs map whole VLAN ranges in advance instead of mapping each VLAN as it is created.