Packet Path/

vPC vs VSS vs StackWise Explained: Two Switches, One Illusion, Three Very Different Brains

A server with two cables plugged into two different switches should survive either switch failing — and should get to use both cables at once. Cisco sells three different ways to arrange that trick, and they look alike in a diagram while behaving very differently on the worst day. vPC keeps two independent Nexus switches that cooperate only at the port-channel. VSS welds two big Catalyst chassis into one logical switch with one brain. StackWise welds a row of smaller Catalyst switches into one logical switch the same way. This article puts all three in one table — control plane, data plane, management, and exactly what a consistency failure, a partner failure, and a broken link between the partners can and cannot do in each — then works three real picks.

Switching · Intermediate · 11 min · November 14, 2026

Network switches on a dark teal background: a pair at the top linked by a glowing dotted line, a large glowing merged chassis in the centre, a stack of four switches circled on the right, and servers connected below.

The short answer

What is the difference between vPC, VSS and StackWise? All three let a downstream device — a server, or another switch — connect to two Cisco switches at once and treat those connections as one bundle, so both links carry traffic and either switch can fail without taking the device offline. They differ in how much the two switches merge. vPC (virtual Port Channel, on Nexus switches) barely merges them at all: each switch keeps its own brain, its own configuration and its own management address, and they cooperate only to present one port channel to the device below. VSS (Virtual Switching System, on classic Catalyst chassis) merges two chassis completely into one logical switch with one brain and one configuration. StackWise (on Catalyst access switches) does the same complete merge for a row of smaller switches joined by special stack cables — up to eight Catalyst 9300 switches in a single stack. One brain is simpler to run and has no partner settings to keep in agreement. Two brains are more work, and need a consistency check, but no single brain failure or upgrade touches both switches at once.

Think of three ways to run a shop with two tills. StackWise and VSS hire one manager for the whole shop: one price list, one set of decisions, tills are just extra serving points — simple, and if the manager's rulebook is wrong, it is wrong everywhere at once. vPC keeps two managers, one per half of the shop, each with their own rulebook, who compare notes on the few rules that must match before they serve a customer together. More notes to compare — and a formal argument, called a consistency check, when the rulebooks disagree — but one manager going home sick never closes the shop, and updating one rulebook never touches the other half.

In plain EnglishControl plane: the switch's brain — the part that learns addresses, runs spanning tree and routing, and decides where traffic should go. Data plane: the hands — the hardware that actually forwards each frame at line speed, following the brain's decisions. Port channel: several physical cables bundled and treated as one logical link, so traffic spreads across them and the bundle survives one cable failing. A bundle that spans two switches is sometimes called a multichassis port channel. Dual-homing: connecting one device to two switches at once, one cable to each. Peer: one partner in a pair. Nothing in this article needs more jargon than that — each technology's own link names are explained where they appear.

The three systems, side by side

QuestionvPCVSSStackWise
Where it livesCisco Nexus switches, mainly data centresClassic Catalyst chassis — the 6500/6800 family is the textbook homeCatalyst access switches, for example the 9300 family
How many switches merge?Exactly 2 peers in a vPC domain — and they do not fully mergeExactly 2 chassisUp to 8 members in a Catalyst 9300 stack
Control planes (brains)2 — one per switch, both active, fully separate1 active for the whole logical switch, with a standby ready to take over1 active for the whole stack, 1 standby, the rest are members
Configuration and management2 configurations, 2 management addresses — you configure each peer1 configuration, 1 management address — you configure the logical switch once1 configuration, 1 management address for the whole stack
The link between partnersA peer link that synchronises state, plus a separate peer-keepalive link that carries only alive-or-dead heartbeatsThe Virtual Switch Link (VSL): ordinary high-speed Ethernet links carrying control and data between the two halves of one switchDedicated stack cables in a ring — StackWise-480 ports on a Catalyst 9300 — forming the stack's backplane
What the device below seesOne port channel, both links activeOne port channel spanning both chassis (a multichassis EtherChannel), both links activeOne switch; a port channel can span members, both links active
If one partner diesThe survivor keeps forwarding on its own links; the device below loses half its bundle, not its connectionThe surviving chassis keeps forwarding; the standby brain takes over the logical switchPorts on the dead member go dark; the rest of the stack keeps forwarding, and the standby becomes active if the dead member was the brain
If the link between partners breaksThe keepalive decides: partner alive but unreachable over the peer link means the secondary peer suspends its vPC ports rather than risk two switches acting aloneThe two halves can both try to run the one logical switch — dual-active — which is why VSS designs add a separate dual-active detection methodOne broken stack cable turns the ring into a chain and the stack keeps working through the remaining path; losing the path entirely splits the stack
Home groundDual-homing servers and switches in a data centreA campus or data-centre core built from two large chassisA wiring closet that needs many ports under one address
vPC — two brains peer link between twoindependent switches VSS — one brain, two chassis one logical switch,one configuration StackWise — one brain, a row stack cables in a ring,up to eight members (9300) Downstream, all three look the same: one bundle, both links working. The difference is entirely in whose brain is in charge — and what can argue, split, or fail.

Read the table from the brain row downwards. The number of control planes decides the configuration row (two brains need two configurations), which decides whether a consistency check can exist at all, which decides a whole family of failure modes. Almost every argument about these three technologies is really an argument about that one row.

One brain: simpler, and the blast radius is the whole switch

VSS and StackWise make the same bet: merge completely. There is one configuration to write, one address to manage, one spanning-tree and routing instance to reason about, and a port channel that spans the pair falls out naturally, because from the network's point of view there is only one switch there. For an access closet or a two-chassis core, that simplicity is not laziness — it is fewer settings to type, fewer settings to mistype, and a design a tired engineer can hold in their head at 3 a.m.

The price is concentration. The one active brain is a single point of decision: a software fault, a bad configuration change, or an upgrade gone wrong applies to the whole logical switch at once, because there is no second, independent switch to carry on while the first is fixed. Standby brains protect against hardware death — if the active switch fails, the standby takes over — but a standby faithfully running the same configuration is not protection against the configuration itself being the problem. And the link between the halves becomes critical in a particular way: if two chassis that believe they are one switch lose contact and both keep deciding, the network gets two copies of the same switch — same addresses, same identity — which is why VSS designs pair the VSL with a dual-active detection method, and why a stack ring is cabled as a ring, so one broken stack cable still leaves a path.

A modern note belongs here, because datasheets have moved on: on current Catalyst 9000 chassis (the 9400, 9500 and 9600 families), the VSS idea is sold as StackWise Virtual — still exactly two switches, still one active brain, but joined by ordinary Ethernet links rather than a classic VSL, so the pair can sit further apart. If a design document says StackWise Virtual where an older one said VSS, it is the same one-brain bet with newer cabling.

Two brains: more work, and a failure stays in its half

vPC makes the opposite bet. Each Nexus switch is a complete, independent switch: its own brain, its own configuration, its own management address, able to route and forward alone. The pair cooperates on exactly one trick — presenting a single port channel to a dual-homed device — and synchronises just the state that trick needs over the peer link. A separate, deliberately simple peer-keepalive link carries nothing but heartbeats; its only job is to answer one question when the peer link fails: is my partner dead, or merely unreachable? If the partner is dead, the survivor carries on alone. If the partner is alive but the peer link is down, the secondary peer suspends its vPC ports — two healthy switches that cannot synchronise must not both act as if they own the shared bundle.

The reward is isolation. One peer can be upgraded, reconfigured, or suffer a software fault while the other keeps forwarding; a mistake typed on one peer does not automatically exist on the other. The cost is exactly the mirror image: because two configurations exist, they can disagree — and some disagreements are dangerous, because the device below believes it is talking to one bundle while the two ends behave differently. That is why vPC, alone of the three, needs the check in the next section.

The consistency check: what it can and cannot do

Before a vPC comes up, and whenever a relevant setting changes, the two Nexus peers compare the settings that have to agree for the shared bundle to behave as one. Cisco sorts those settings into two severities. A Type-1 mismatch is a setting that would break forwarding or risk a loop if the peers disagreed — with the default graceful behaviour, the primary peer keeps its vPC ports up and the secondary peer brings its vPC ports down, so the dual-homed device keeps working through one switch instead of being served inconsistently by two. A Type-2 mismatch generates a warning and the vPC stays up — forwarding continues, which is sneakier, because everything looks healthy while a quieter setting differs between the ends.

Two honest limits matter. First, the check can only compare settings it knows about, on the two peers — it cannot tell you the design itself is wrong, and it says nothing about the device below. Second, the equivalent failure simply does not exist in VSS or StackWise, not because they solved it, but because there is no second configuration to compare: a wrong setting in a merged system is not a mismatch, it is just the configuration, live everywhere at once. When the symptom in front of you is a failed vPC consistency check — the peer states, the exact commands in order, and the safe fix — work through the companion case: vPC consistency check failed. And when a port channel refuses to bundle at all, the log line is usually a different one — see the decoder entry for %EC-5-CANNOT_BUNDLE2.

Three worked picks

Pick one — a wiring closet serving a floor of desks and phones. The job: a few hundred access ports in one closet, one address to manage, uplinks that survive one switch dying. This is StackWise's home ground and it is not close: stack the Catalyst switches with their stack cables, configure the stack once, spread the uplinks across members. Reaching for vPC here buys two-brain complexity on a layer that gains nothing from it — and Nexus switches do not belong in a wiring closet in the first place.

Pick two — servers dual-homed in a data centre. The job: each server or top-of-rack switch connects to two aggregation switches, both links active, and either aggregation switch must be upgradeable without an outage window for the whole rack. This is vPC's home ground: two independent Nexus peers, a peer link and keepalive between them, one vPC per dual-homed device. The two-brain design is precisely what makes the one-at-a-time upgrade possible, and the consistency check is the price of admission — budget the operational care it asks for, matching Type-1 settings on both peers every time.

Pick three — a campus core of two large chassis. The job: two big chassis in the core, one logical core switch, port channels down to every distribution block. Classic VSS was built for exactly this, and on modern Catalyst 9000 chassis the same design is StackWise Virtual. The one-brain simplicity suits a core whose configuration changes rarely and deliberately — with dual-active detection designed in from the start, not bolted on after the first VSL scare.

The decision, in four questions, and the mistakes

1. Which layer are you building? Access closet: StackWise. Data-centre dual-homing: vPC. Two-chassis core: VSS, or StackWise Virtual on current hardware. The platforms rarely even overlap — a Nexus does not stack, and a Catalyst access switch does not run vPC.

2. Must one partner survive the other's upgrade and mistakes, or is one configuration worth more? Independent failure and upgrade domains point to two brains. One configuration, one address, and the simplest possible mental model point to one brain.

3. What breaks if the link between the partners breaks? Answer this on paper before buying anything: vPC answers it with the keepalive and secondary suspension, VSS answers it with dual-active detection, a stack answers it with the ring. A design with no answer to this question is not finished.

4. Who keeps two configurations identical, forever? If the honest answer is nobody, a two-brain design will eventually hand you a Type-1 suspension at an inconvenient hour — or worse, a Type-2 warning nobody read. Process is part of the product.

The mistakes to avoid. Describing vPC peers as one switch in documentation — the next engineer will configure one peer and wonder why the bundle is suspended. Treating a Type-2 warning as noise because traffic still flows. Building VSS without dual-active detection because the VSL has two links and feels safe. Breaking a stack ring to borrow a cable. And comparing these technologies by port-channel behaviour alone — downstream they are deliberately identical; the choice is entirely about brains, blast radius, and operations.

Fact-check sourcesvPC peer link, peer-keepalive (heartbeat only, no data or synchronisation traffic, 1-second default interval), vPC domain composition, and Type-1 versus Type-2 consistency behaviour — including the default graceful check, where the primary keeps the vPC up and the secondary brings it down on a Type-1 mismatch, and Type-2 generating a warning while the vPC stays up — follow Cisco's Nexus vPC operations and interfaces configuration guides. StackWise member counts and roles — a Catalyst 9300 stack has up to eight members through StackWise-480 ports, with one active and one standby switch and the active as the single point of stack-wide management — follow Cisco's Catalyst 9300 stacking and high-availability configuration guide (IOS XE 17.13.x). VSS as two Catalyst chassis forming one logical switch with one active control plane over a Virtual Switch Link, and StackWise Virtual as its two-switch successor on Catalyst 9400/9500/9600 over Ethernet links, follow Cisco's VSS and Catalyst 9000 StackWise Virtual documentation. Exact platform support varies by model and software release — a production design verifies its own pair against the chosen platform's current guide.

Try it yourself

Take any two-switch design you know and audit it the way this article reads it: list the settings the two partners must agree on, then check a real pair of configurations for the mismatches a consistency check would catch — MTU, VLAN and port-channel disagreements included — with the free network config verifier. It runs entirely in your browser; paste the two configs, read the findings, and ask of each one: in a one-brain system, would this have been a mismatch at all — or just the live configuration?

Key takeaways

Keep reading