Catalyst 9300 StackWise vs StackWise Virtual: Deployment Guide

2026年9月2日 LOG admin

Choosing between Cisco StackWise and StackWise Virtual is really a question about where the switch sits in the network. Physical StackWise targets the access layer - up to eight Catalyst 9300s joined by proprietary copper cables inside one rack - while StackWise Virtual is the core/distribution-layer technology that binds exactly two Catalyst 9500/9600 chassis over standard fiber. This article explains the architecture differences, the platform and licensing matrix, and the practical constraints behind each choice.

The confusion is understandable, because both technologies present the operator with the same abstraction: several physical switches that behave as one logical device with a single management address, a single control plane and a single spanning-tree instance. The similarity ends there. StackWise extends a chassis over a short proprietary backplane; StackWise Virtual makes two independent chassis cooperate across a standards-based link. Those two sentences drive every downstream difference in distance limits, optics, licensing, power design and failure behaviour.

StackWise: Physical Access-Layer Stacking

StackWise connects Catalyst 9200/9300 switches into a closed-loop ring with a shared control plane and one management IP. The backplane has evolved with the Catalyst 9000 family: StackWise-160 on Catalyst 9200 (160 Gbps), StackWise-480 on Catalyst 9300, StackWise-1T on Catalyst 9300X, and the next-generation 1.6 Tbps on Catalyst 9350. Up to eight or nine switches (platform dependent) can stack, and StackPower lets power supplies be pooled across members.

The ring topology matters operationally. Because the stack cables form a loop, a single cable failure does not break the stack — traffic reroutes around the ring and the stack logs a partial-ring condition. Lose two cables and the stack splits, producing two independent forwarding domains that usually generate duplicate IP address alarms, duplicate spanning-tree roots or a renegotiated active switch. Designing racks so that the ring cables never share a conduit or a patch panel with the uplinks is the cheapest insurance against a split.

Switch# show switch
Switch# show switch stack-ports summary
Switch# show switch detail
Switch# show stack-power

Active-switch election is deterministic: the switch with the highest priority wins, with MAC address as the tie-breaker, and a stack that has already elected an active will not preempt unless you explicitly request re-election during a maintenance window.

StackWise Virtual: Two Chassis, One Logical Switch

StackWise Virtual (SVL) is the modern successor to VSS, rebuilt natively for IOS-XE. Exactly two switches are combined over standard 10G/40G/100G fiber links - meaning the pair can be kilometers apart instead of 3 meters apart. The active switch runs the control plane for both chassis and encapsulates control traffic with a Virtual Switch Header (VSH) across the SVL to the standby. Because it uses standard optics, SVL also enables geographic redundancy across buildings or data centers.

Two design details separate SVL from any form of physical stacking. First, the links are ordinary Ethernet ports running a proprietary encapsulation, so they traverse DWDM, dark fibre or a campus backbone exactly like any other 100G service — subject to the latency budget IOS-XE enforces on the virtual link. Second, the pair keeps a dual-active detection mechanism, historically a fast-hello link over a direct connection through a non-virtualised intermediate device. Without a working dual-active detection path, a split SVL pair will both believe they are active, and both will start answering for the same virtual MAC — the single worst failure mode in the design.

Switch(config)# stackwise-virtual
Switch(config-stackwise-virtual)# domain 1
Switch(config-stackwise-virtual)# exit
Switch(config)# interface range FortyGigabitEthernet 1/0/1-2
Switch(config-if-range)# stackwise-virtual link 1
Switch(config-if-range)# exit
Switch(config)# stackwise-virtual dual-active-detection
Switch(config-stackwise-virtual-dad)# exit

Verify with show stackwise-virtual, show stackwise-virtual link and show stackwise-virtual dual-active-detection. The link count and the DAD state are the two fields worth alerting on in production.

Platform Support and Licensing Matrix

Catalyst 9200 - physical StackWise-160 only.
Catalyst 9300 / 9300X - native physical StackWise-480/1T; a 9300 pair can run StackWise Virtual only in collapsed-core designs, and you cannot run both modes at once.
Catalyst 9350 - next-gen 1.6 Tbps physical stacking.
Catalyst 9400 / 9500 / 9600 - StackWise Virtual only; no proprietary stacking ports.

StackWise Virtual requires the Cisco DNA Advantage (or Catalyst Advantage) license - the DNA Essentials tier does not include SVL, a common procurement gotcha. The licensing trap is worth restating in project language: a design that depends on SVL can be fully racked, cabled and technically functional while remaining non-compliant because the Smart Licensing tier was ordered at Essentials. Because IOS-XE enforces feature entitlement rather than merely reporting it, the configuration will simply refuse to activate, and the error appears when someone tries to enable the feature rather than when the purchase order is cut.

Key Differences at a Glance

Feature	Physical StackWise	StackWise Virtual
Deployment tier	Access	Core & distribution
Maximum members	Up to 8-9	Exactly 2
Cabling	Proprietary copper stack cables	Standard 10G/40G/100G fiber
Distance	~3 m (single rack)	Kilometers
Power pooling	Yes (StackPower)	No

Cabling, Optical Budget and Distance Rules

Because SVL runs over standard optics, the physical design questions shift from "does the cable reach across the rack?" to something closer to a WAN circuit provision. Each link must meet the receive-sensitivity and dispersion budget of the installed transceiver, and the pair must remain within the supported latency for the virtual link — typically a small number of milliseconds round trip. Crossing a heavily loaded provider network or a long chain of intermediate switches is the usual reason a proposed SVL design is rejected during review.

Use at least two physical links between the chassis, and preferably three or four spread across different line cards where the platform allows it. A single fibre pair carrying the entire control plane is a single point of failure with extra steps. On the physical stacking side the equivalent guidance is to keep the ring closed, use the supported cable lengths for the model, and never mix cable generations across a single ring unless the platform documentation explicitly permits it.

Failure Behaviour and Dual-Active Detection

The two technologies also differ in how they fail, which matters more than how they run. In a physical stack, member failure is graceful: the stack re-elects, the remaining members forward, and management of the survivors is unaffected. In an SVL pair, the failure of the active chassis triggers a standby takeover that depends on the virtual link and the DAD link both being healthy to complete cleanly.

Dual-active detection deserves particular attention during maintenance. Any change that removes the DAD path while the virtual link is also unstable creates the exact condition the mechanism exists to prevent. Sequence maintenance so that at most one of the two paths is disturbed at a time, and verify DAD state after every change rather than assuming it survived the window.

Design Guidance: Which One Where

The decision usually resolves in one pass once the tier is fixed. If the goal is a wiring-closet stack of access switches with a single management address and pooled power, physical StackWise is the correct and cheaper answer, and SVL adds cost without adding value. If the goal is a distribution pair or a collapsed-core design where the two switches must survive the loss of a building, an electrical feed or an entire rack, only StackWise Virtual delivers that separation.

Watch for the mixed requirement, which is where most mistakes occur. An organisation that wants "high availability everywhere" may ask for SVL at the access layer, not realising that the Catalyst 9300 supports it only in a collapsed-core role and that the DNA Advantage licence is required. Equally, a distribution design specified with physical stacking will be physically impossible on a Catalyst 9500, which has no stacking ports at all. Establish the switch model and the licence tier first, then choose the technology; reversing that order produces designs that cannot be built.

Uplink and Topology Design Around the Virtual Pair

Once two chassis present as one logical switch, every downstream device must treat the pair as a single neighbour, and that changes the uplink design. For an SVL distribution pair, downstream access switches should be dual-homed with one link into each chassis and configured with a port channel; because both chassis share a control plane and appear as one spanning-tree root, the standard EtherChannel logic across the pair works without any additional protocol. Leaving a switch single-homed to one chassis silently removes the redundancy the pair was purchased to provide.

The same logic applies upwards. An uplink to a core should be a port channel with member links landing on both chassis. Designers who cable all uplinks to a single member recreate a single point of failure and, worse, an asymmetric traffic pattern that can look like a software bug when a failover test produces a traffic loop or a flood.

Management, Operations and Software Upgrades

One management address is the headline benefit and also the main operational trap. Both a physical stack and an SVL pair issue their software image from a single active, so an upgrade is a single coordinated event rather than a per-device rollout. Plan the reload sequence explicitly: with SVL, the active and standby must be upgraded in order with the virtual link state verified between steps, and with a physical stack, members reload one at a time in a defined order so that the stack never loses its active mid-upgrade.

Automation and monitoring need the same care. SNMP polling, syslog sources and NetFlow collectors all see one logical device, which simplifies inventory but means a per-member hardware fault can hide behind a healthy aggregate. Poll per-member hardware sensors where the platform exposes them, and alert on stack port and virtual link counters rather than only on device reachability.

Switch# show version | include Model|SW Version
Switch# show switch stack-ports summary
Switch# show stackwise-virtual link
Switch# show redundancy

Migration and Coexistence

Converting an existing deployment is rarely a single change. A stack of Catalyst 9300s cannot be half-migrated to SVL, because the two technologies are mutually exclusive on the same chassis pair — a 9300 pair runs either physical stacking or StackWise Virtual, never both. The practical path is to build the target pair as a separate logical device, move service by service while both topologies exist in the same network, and remove the old stack only once the last subnet has been migrated and the spanning-tree topology has been verified as stable.

During coexistence, keep the Layer 2 domains distinct or use routing to separate them. Two logical devices announcing the same virtual MAC or the same HSRP address in one broadcast domain is an outage waiting for a maintenance window, and duplicate-address alarms during a migration are far more often a design mistake than a hardware fault.