network · deep · draft

Collapsed core

A two-tier campus where distribution and core roles share one redundant aggregation pair. Access dual-homes north; WAN and shared services attach at that pair — there is no separate core tier.

In one sentence. A full three-tier design adds a dedicated core pair that is unnecessary when the site is small and distribution can also perform campus aggregation and edge/services attachment.

Why it exists

Cisco Campus LAN/WLAN CVDs describe collapsing distribution and core for smaller sites (often one building or a few adjacent buildings with adequate fiber). The separate core tier is omitted so distribution becomes the Layer-3 aggregation point for access and often for WAN/Internet-edge/services connectivity.

Visual walkthrough

Walkthrough

Two tiers: access into a collapsed core

Distribution and core roles share one redundant pair. Access dual-homes north; WAN and shared services attach at the collapsed core — there is no separate core tier.

Step 1 / 4
WAN / Internet edgeFirewall / servicesCC1 · collapsed coreCC2 · collapsed coreA1 · accessA2 · accessHost · VLAN 10Host · VLAN 20

Access switches attach endpoints. The collapsed-core pair aggregates those uplinks, hosts Layer-3 gateways, and connects toward WAN, firewall, and shared services — the separate core tier is omitted on purpose for a small site.

Walkthrough

East–west inter-VLAN via the collapsed core

Hosts in different VLANs meet at Layer-3 SVIs on the collapsed core. Same-VLAN flooding stays local when the L2 design allows; cross-VLAN always climbs to the CC pair.

Step 1 / 4
WAN / Internet edgeFirewall / servicesCC1 · collapsed coreCC2 · collapsed coreA1 · accessA2 · accessHost · VLAN 10VLAN 10Host · VLAN 20VLAN 20

Host on VLAN 10 and host on VLAN 20 share the campus fabric but not a broadcast domain. Inter-VLAN routing is concentrated on the collapsed-core gateways.

Exact IRB/SVI placement and FHRP pairing are site-specific; this walkthrough is illustrative of the common collapsed-core pattern.

Walkthrough

One collapsed-core member fails

Losing one CC member reduces capacity and HA paths. Dual-homed access should keep forwarding on the surviving member when uplinks and FHRP/routing are designed for that failure.

Step 1 / 4
WAN / Internet edgeFirewall / servicesCC1 · collapsed coreCC2 · collapsed coreA1 · accessA2 · accessHost · VLAN 10Host · VLAN 20

Both collapsed-core members are forwarding. Access uplinks and services attachments are dual-homed. Capacity and change windows still couple routing and often WAN attachment on this pair.

Control vs data plane

Control plane

FHRP and campus routing typically live on the collapsed-core pair. Access runs STP/LACP as designed. Exact hello timers, FHRP preferences, and WAN peering defaults are site-specific and are not asserted as universal values on this page.

Data plane

Access switches attach endpoints. Northbound uplinks dual-home into the collapsed-core pair. Inter-VLAN routing concentrates on collapsed-core SVIs/IRB. WAN, firewall, and shared services often attach on or near that same pair.

Request / packet path

North–south

Endpoint → access → collapsed core → firewall/WAN/services. Services co-located on the collapsed core mean edge attachment and user aggregation share a change domain.

East–west

Inter-VLAN east-west climbs to the collapsed core for routing, then returns to access. Same-VLAN forwarding depends on whether L2 is local at access or extended — that choice is design-specific (unknown as a universal default).

Scaling & math

Without a distinct distribution↔core hop, oversubscription planning collapses to the access↔aggregation relationship. Cisco CVDs treat ratios as business-driven and note that historical implementations have used access↔aggregation oversubscription up to 20:1 — an observed range, not a mandate. Exact maximum access switches per collapsed-core platform is platform-specific (unknown here).

When it breaks

  • One collapsed-core member is down or drained; dual-homed closets should stay up with reduced HA capacity.

    Cause. Single CC chassis/stack member failure or maintenance.

    Mitigation. Confirm access dual-homing and FHRP/routing adjacency on the surviving member before maintenance; size the pair for acceptable capacity loss.

  • Site-wide L3 aggregation and typically WAN attachment fail together.

    Cause. Both collapsed-core members unavailable, or a shared control-plane fault spanning the pair.

    Mitigation. Treat the pair as a single change/fault domain for major work; prefer designs with stronger isolation when that blast radius is unacceptable.

  • A maintenance window on aggregation also risks WAN/services co-located on the same pair.

    Cause. Weaker change isolation than a dedicated core that separates user aggregation from edge services.

    Mitigation. Schedule dual impacts deliberately, or keep WAN/services on a more isolated attachment when the CVD recommends a dedicated core.

  • Large STP/broadcast domain; slow convergence or instability across many access closets.

    Cause. Layer-2 stretched from many access switches into the collapsed core without segmentation.

    Mitigation. Limit L2 domains; prefer L3 at the collapsed core with smaller access VLANs where the design allows.

  • Loss or latency under load even though the topology is “up”.

    Cause. Access↔aggregation uplink capacity oversubscribed relative to demand — a planning choice called out in CVD language (including historical examples up to 20:1).

    Mitigation. Recalculate access↔aggregation oversubscription against business requirements; add uplinks or reduce contention.

Misconceptions

  • “Collapsed core means a single non-redundant switch.” — CVD language describes a redundant collapsed-core / distribution pair (or logical pair), not a lone box.
  • “Always prefer three-tier; collapsed core is outdated.” — Cisco documents collapsed core as appropriate for smaller sites and calls out geographic dispersion / fiber cost as reasons to keep a dedicated core.
  • “Access↔aggregation oversubscription is fixed at 20:1.” — CVDs treat ratios as business-driven; “up to 20:1” is an implemented historical range, not a universal default.

Reference expression

Primary teaching sources are Cisco Campus LAN and Wireless LAN CVD language, Enterprise Campus 3.0 architecture overview (when not to stay collapsed), and the Aug 2013 Campus Wired LAN configuration files guide for two-tier collapsed-core wording. Short excerpts only — follow the linked sources for full validated designs. review: draft.

Standards & sources

Known unknowns

  • Exact maximum access switches or ports per collapsed-core platform are platform-specific and not tabulated here.
  • FHRP timers, STP mode defaults, and WAN peering details vary by vendor OS and site policy — not asserted as universal defaults.

Check yourself

  1. What does “collapsed core” omit compared with classic three-tier campus?

    • The access tier
    • A separate dedicated core tier
    • All Layer-3 routing
    • WAN attachment

    Answer: A separate dedicated core tier. Distribution takes on core aggregation roles so a separate core pair is not required for that site size.

  2. Where do access switches typically dual-home in this pattern?

    • Only to the WAN firewall
    • To both members of the collapsed-core pair
    • Only to other access switches
    • Directly to the Internet

    Answer: To both members of the collapsed-core pair. Dual-homing into the CC pair is how a single uplink/CC member failure is survived.

  3. Why can maintenance on the collapsed core be riskier than on a dedicated core?

    • Because STP cannot run at access
    • Because user aggregation and often WAN/services share the same pair’s change domain
    • Because dual-homing is forbidden
    • Because oversubscription is always 1:1

    Answer: Because user aggregation and often WAN/services share the same pair’s change domain. Enterprise Campus guidance highlights weaker change isolation when edge and aggregation share the collapsed pair.

  4. Cisco CVD language on access↔aggregation oversubscription treats 20:1 as…

    • A mandatory default for every campus
    • A business-driven ratio, with historical implementations up to 20:1 noted
    • Impossible on copper
    • Only valid for wireless

    Answer: A business-driven ratio, with historical implementations up to 20:1 noted. Ratios are business-driven; “up to 20:1” is documented as implemented, not mandated.

  5. When does Cisco guidance push away from staying collapsed?

    • Any site with more than one VLAN
    • Geographic dispersion with many buildings and long fiber runs raising aggregation cost
    • Whenever OSPF is used
    • Whenever hosts are dual-homed

    Answer: Geographic dispersion with many buildings and long fiber runs raising aggregation cost. Fiber cost and aggregation fan-in across dispersed buildings are cited reasons for a dedicated core.

  6. Inter-VLAN east-west in this teaching topology typically…

    • Stays only on the access switch with no gateway
    • Climbs to the collapsed core for routing, then returns to access
    • Requires a spine tier
    • Must hairpin through the public Internet

    Answer: Climbs to the collapsed core for routing, then returns to access. L3 gateways concentrate on the collapsed core in the two-tier pattern.

  7. Is collapsed core the right class for a large east-west DC fabric?

    • Yes — it replaces Clos
    • No — DC east-west scale is a different architecture class (e.g. spine-leaf)
    • Yes — if you add more VLANs
    • Only with VXLAN

    Answer: No — DC east-west scale is a different architecture class (e.g. spine-leaf). Campus collapsed core is not a substitute for DC Clos fabrics.

  8. Primary sources used on this page include…

    • Anonymous forum posts only
    • Cisco Campus LAN/WLAN CVD and Enterprise Campus 3.0 overview materials
    • Invented show-command captures
    • Unlabeled AI summaries

    Answer: Cisco Campus LAN/WLAN CVD and Enterprise Campus 3.0 overview materials. Claims are tied to CVD / campus architecture primary docs with provenance.