Enhanced Layer-2 Non-Blocking Multipath Fabric Architecture

A Conceptual Multipath Layer-2 Fabric Architecture with Switching Fabric Segmentation for Datacenter and Enterprise Networks

Conceptual Technical Design Paper

Author: Christopher Soans
Date: October 2026

Download PDF

VEDI Enhanced Layer-2 Fabric Architecture
Figure 1. VEDI Enhanced Layer-2 Fabric Architecture

Abstract

Traditional Ethernet Layer-2 networks rely heavily on Spanning Tree Protocol variants such as STP, RSTP, and MSTP to prevent forwarding loops. While effective, spanning-tree designs intentionally block redundant links and organize the network around a root-bridge topology. Modern alternatives such as IEEE 802.1aq Shortest Path Bridging demonstrate that Layer-2 networks can use topology-aware shortest-path forwarding and keep multiple physical paths active.

This paper proposes a conceptual enhanced Layer-2 fabric architecture based on Virtual Endpoint Discovery Interface (VEDI) and a distributed Endpoint Binding Database (EBD). The architecture preserves standard Ethernet mechanisms wherever practical while replacing the spanning-tree forwarding model with authenticated endpoint distribution, explicit Fabric Switch identities, shortest-path/multipath forwarding, optional switch-to-switch Fabric Congestion Notification (FCN), FSID/VLAN-aware EBD propagation, and intelligent handling of endpoint mobility, multihoming, failures, broadcasts, multicast traffic, virtualization, wireless access, and other trunked endpoint devices. The first-generation data plane reuses IEEE 802.1ad Q-in-Q between Fabric Switches, with the 12-bit S-VID directly representing the FSID and the inner C-VID preserving the endpoint VLAN.

The architecture does not propose replacing Ethernet, IEEE 802.1Q VLANs, IEEE 802.1ad Q-in-Q, LAG/LACP, vPC/MLAG, IGMP/MLD, or other useful Layer-2 mechanisms. Its primary architectural change is the replacement of STP/RSTP/MSTP root-bridge election and topology blocking inside the enhanced fabric. A future native 24-bit FSID header is considered an optional evolution path rather than a first-generation requirement.

The proposal remains conceptual and is intended for further simulation, standards analysis, implementation research, and prototype validation.

1. Introduction

Layer-2 Ethernet remains one of the most widely deployed networking technologies because of its simplicity, interoperability, hardware efficiency, and broad support across servers, switches, storage systems, virtualization platforms, wireless systems, and edge devices.

The fundamental problem with conventional switched Ethernet is not Ethernet framing itself. The difficulty arises when redundant physical paths exist. Traditional STP-family protocols prevent loops by reducing the active forwarding topology to a loop-free tree. This approach is reliable but creates several limitations:

  • redundant links may remain blocked during normal operation;
  • forwarding topology is influenced by root-bridge placement;
  • topology convergence can temporarily disrupt traffic;
  • endpoint discovery relies heavily on passive MAC learning;
  • unknown traffic may be flooded;
  • MAC aging can cause previously known endpoints to become unknown;
  • active-active forwarding across the entire Layer-2 fabric is limited.

Technologies such as Shortest Path Bridging (SPB) demonstrate that these limitations are not inherent to Ethernet. A Layer-2 fabric can use link-state topology information and shortest-path forwarding instead of a spanning tree. The architecture proposed here builds on that principle while adding a stronger endpoint-intelligence layer.

2. Design Objective

The primary objective is to preserve native Ethernet and existing useful Layer-2 mechanisms while replacing spanning-tree-based topology construction with validated endpoint intelligence and active multipath Layer-2 forwarding.

The architecture is intended primarily for greenfield deployments. Legacy STP/RSTP/MSTP environments should remain operationally separate and connect to the enhanced fabric through a Layer-3 boundary. This avoids mixing two fundamentally different loop-prevention models inside the same Layer-2 domain.

3. Existing Technology as the Foundation

3.1 IEEE 802.1Q

Standard IEEE 802.1Q tagging remains unchanged on endpoint-facing access and trunk interfaces. In the first-generation enhanced fabric, IEEE 802.1ad Q-in-Q is reused between Fabric Switches so that the outer Service VLAN identifier (S-VID) represents the Fabric Segment Identifier (FSID) and the inner Customer VLAN identifier (C-VID) preserves the endpoint VLAN. This provides a practical migration path without requiring a new wire protocol in the initial implementation.

3.2 LAG and LACP

Link aggregation continues operating normally. LACP provides valuable information including logical LAG membership, synchronized member state, collecting/distributing state, and physical member availability. VEDI consumes this information rather than replacing it.

3.3 vPC and MLAG

vPC/MLAG-style paired-switch designs remain supported. The enhanced fabric adds explicit endpoint provenance, independent physical switch identities, EBD synchronization, degraded-pair signaling, and independent endpoint validation during split-brain conditions.

3.4 IGMP and MLD Snooping

Existing IPv4 IGMP and IPv6 MLD mechanisms remain the primary means of learning multicast receiver membership. VEDI correlates that information with validated endpoint identities and distributes multicast reachability more intelligently through the fabric.

3.5 Shortest Path Bridging

SPB provides an important technical precedent. It demonstrates that Ethernet forwarding does not require a spanning tree, redundant links may remain active, link-state topology information can drive Layer-2 forwarding, and shortest paths and ECMP-like behavior are feasible at Layer 2. This architecture therefore does not claim that shortest-path Layer-2 forwarding itself is new. The new emphasis is on combining shortest-path forwarding with validated endpoint ownership and distributed endpoint intelligence. The enhanced fabric similarly reuses mature link-state concepts for topology synchronization and shortest-path computation, but abstracts them behind native Fabric Domain operation so that operators do not configure IS-IS or another routing protocol separately.

4. Fabric Switch Identifier (FSW-ID)

Every enhanced Fabric Switch has a unique FSW-ID derived from the switch's persistent system or base MAC address rather than being manually assigned.

FSW-ID: 00:1B:54:7A:92:10

The FSW-ID is used for endpoint origin identification, EBD advertisements, topology information, path-loop detection, paired-switch membership, failure reporting, and EBD provenance. Duplicate active FSW-IDs within the same Fabric Domain are treated as a critical error.

5. Fabric Domain

All enhanced switches belong to a configured Fabric Domain. VEDI is enabled at the Fabric Domain level rather than independently on every VLAN.

fabric-domain AI-FABRIC
  authentication key-id 1
  authentication password ********
  vedi enable

Participating VLANs inherit VEDI capability. A per-VLAN override such as no vedi may be supported for exceptional cases.

6. Fabric Uplink Adjacency

Production Fabric Uplinks should require explicit fabric-mode configuration. Unlike ordinary endpoint trunks, a Fabric Uplink may transport multiple FSIDs and uses Q-in-Q for data traffic. The operator controls which FSID/VLAN combinations are permitted across the adjacency.

interface Ethernet1/49
  switchport mode fabric
  switchport fabric control vlan 999
  switchport fabric allowed vlan 120,130,140 fsid 1250
  switchport fabric allowed vlan 120,130,140 fsid 1251
  switchport fabric allowed vlan 10,20,30 fsid 1252

An optional auto-negotiation mode may be provided for smaller deployments, but a link must still complete fabric authentication, capability checks, MTU validation, FSID/S-VID compatibility checks, topology synchronization, and required EBD synchronization before becoming ACTIVE.

A fabric adjacency becomes active only after:

  1. physical link establishment;
  2. fabric-neighbor discovery;
  3. Fabric Domain comparison;
  4. shared-secret authentication;
  5. FSW-ID exchange;
  6. protocol-version and capability negotiation;
  7. automatic Layer-2 MTU provisioning and peer MTU compatibility validation;
  8. topology synchronization;
  9. required EBD synchronization.

7. Fabric Authentication

The baseline design uses a Fabric Domain shared secret rather than PKI. The password itself is never transmitted. Challenge-response authentication using an HMAC-style mechanism should validate fabric membership. Key IDs should allow controlled password rotation.

This model authenticates membership in the Fabric Domain rather than providing strong cryptographic proof of individual hardware identity. Stronger authentication may be added later as an optional extension.

8. Fabric Control and Q-in-Q Fabric Transport

Fabric hello, authentication, capability, EBD, and topology-control communication uses a designated native/control VLAN on Fabric Uplinks. The control VLAN is separate from Q-in-Q user data and must never be pruned from an active Fabric Uplink. Loss of the required control VLAN prevents fabric adjacency activation.

FSID/VLAN Topology and Configuration Example
Figure 2. FSID/VLAN Topology and Configuration Example

8.1 Fabric Segment Identifier (FSID)

The first-generation architecture defines FSID as a 12-bit Fabric Segment identifier represented directly by the IEEE 802.1ad S-VID. Usable FSID values therefore follow the usable S-VID range of 1 through 4094. The direct identity mapping is intentional: FSID 1250 is transported as S-VID 1250, eliminating adjacency-local translation tables and making packet captures, troubleshooting, and configuration easier to understand.

FSID 1250  =  S-VID 1250
FSID 1251  =  S-VID 1251
C-VID 120  =  endpoint VLAN 120

The FSID defines the logical fabric or virtual-switch namespace. The C-VID defines the VLAN/broadcast domain within that FSID. Consequently, FSID 1250 / VLAN 120 and FSID 1251 / VLAN 120 are independent Layer-2 domains even when they exist on the same physical switch.

8.2 Per-FSID VLAN Namespace

Each FSID maintains its own logical VLAN context, EBD state, multicast state, broadcast state, and forwarding policy. The same 802.1Q VLAN number may therefore be reused in different FSIDs on the same switch without creating a collision. Internally, the authoritative context is FSID + VLAN.

fabric-segment 1250
  vlan 120
    name APP-A

fabric-segment 1251
  vlan 120
    name APP-B

The configuration model keeps familiar VLAN concepts but makes the Fabric Segment the parent configuration context. This prevents a repeated VLAN number from being interpreted as the same global VLAN object and allows the implementation to treat each FSID/VLAN combination as a distinct logical bridge-domain object.

8.3 Endpoint-Facing Interfaces

In the initial implementation, each endpoint-facing physical or logical interface belongs to exactly one FSID. All access or trunk VLANs permitted on that interface operate within that FSID namespace. Endpoints continue using standard 802.1Q and do not need to understand FSID or Q-in-Q.

interface Ethernet1/10
  switchport mode trunk
  switchport trunk allowed vlan 120,130,140
  fabric-segment 1250

interface Ethernet1/11
  switchport mode trunk
  switchport trunk allowed vlan 120,130,140
  fabric-segment 1251

Thus VLAN 120 received on Ethernet1/10 belongs to FSID 1250, while VLAN 120 received on Ethernet1/11 belongs to FSID 1251. This permits overlapping VLAN numbering across independent server, tenant, application, or administrative environments while retaining ordinary 802.1Q behavior at the endpoint.

8.4 Fabric-Link Pruning

Fabric Uplinks may carry multiple FSIDs. Pruning is expressed as permitted FSID/VLAN combinations so that both data-plane forwarding and EBD/control-state propagation are constrained by the same policy.

interface Ethernet1/49
  switchport mode fabric
  switchport fabric allowed vlan 120,130,140 fsid 1250
  switchport fabric allowed vlan 120,130,140 fsid 1251
  switchport fabric allowed vlan 10,20,30 fsid 1252

Because the S-VID directly represents FSID, no dynamic S-VID-to-FSID translation is required in the first-generation architecture. A particular fabric link can carry only the FSIDs required across that adjacency, making the approximately 4094 usable S-VID values a per-link transport resource rather than a requirement that every link carry every segment in the datacenter.

8.5 Automatic Fabric MTU Handling

Q-in-Q adds an additional VLAN tag to fabric data frames. Fabric mode therefore automatically provisions sufficient Layer-2 MTU headroom and advertises supported and required frame size during capability negotiation. A Fabric Uplink must not transition to ACTIVE if either endpoint cannot support the required frame size. This avoids manual MTU mismatches that could otherwise create selective packet loss.

Link Up → Discovery → Authentication → Capability Exchange → MTU Validation → Topology/EBD Synchronization → ACTIVE

8.6 Future Native 24-Bit FSID Evolution

Q-in-Q is the preferred first implementation because it reuses established Ethernet mechanisms and keeps the proposal focused on replacing spanning-tree topology control. If future scale requires more than the 12-bit FSID space available on a fabric link, a later enhanced version may define a native Layer-2 fabric header containing a 24-bit FSID. That would provide approximately 16.7 million FSID values while preserving the same logical FSID + VLAN + MAC EBD model. The native 24-bit header is therefore an evolution path, not a prerequisite for the initial architecture.

9. Virtual Endpoint Discovery Interface

VEDI replaces passive MAC learning as the primary endpoint-authority mechanism. VEDI maintains validated relationships between FSID, VLAN, MAC address, IP address or addresses, local physical port, LAG or logical attachment, Origin FSW-ID, endpoint state, and other local attachment metadata.

FSID:             1250
VLAN:             120
MAC:              AA:BB:CC:DD:EE:FF
IP:               10.120.1.25
Origin FSW-ID:    00:1B:54:7A:92:10
Attachment Type:  LOCAL-PHYSICAL
Interface:        Ethernet1/7
State:            VERIFIED

10. FSID- and VLAN-Aware Endpoint Identity

The authoritative Layer-2 endpoint key is FSID + VLAN + MAC address rather than MAC address alone. The FSID provides the independent logical switching namespace, while the VLAN identifies the broadcast domain within that namespace. Overlapping VLAN numbers are supported across different FSIDs, but deliberate duplicate MAC addressing is not recommended as a normal deployment model.

FSID 1250 / VLAN 120 / MAC-AA → Port 7
FSID 1251 / VLAN 120 / MAC-BB → Port 12

These are independent endpoint bindings. If the same MAC is observed at more than one location within the same FSID/VLAN, the new observation must first be locally verified by VEDI before any EBD ownership change is accepted. If both locations remain independently valid and are not part of an authorized multihoming relationship, the binding enters a duplicate-conflict state rather than oscillating between locations.

11. EBD Propagation

VEDI distributes endpoint state to neighboring Fabric Switches. A new endpoint update contains information such as FSID, VLAN, MAC, IP, Origin FSW-ID, endpoint state, sequence/generation, logical attachment information, path information, and accumulated Layer-2 path cost where applicable.

Remote switches do not actively query endpoints they do not own. Local VEDI validates local attachments; remote endpoint state is learned from authenticated EBD advertisements. This prevents VEDI from becoming a distributed ARP generator.

12. Selective Incremental EBD Synchronization

EBD propagation is incremental. If a neighbor is synchronized through update 5,000 and the local database later reaches update 15,000, only updates 5,001 through 15,000 are transmitted.

Each neighbor maintains synchronization information such as the last acknowledged EBD sequence, current EBD epoch, pending update range, and synchronization state. Updates may be batched into bounded control messages.

Full EBD synchronization is required only when establishing a new adjacency, synchronization consistency is lost, required journal history has expired, or recovery explicitly requires complete resynchronization.

13. FSID/VLAN-Aware EBD Filtering

Fabric-link pruning constrains EBD propagation by FSID and VLAN. If a Fabric Uplink permits VLANs 120, 130, and 140 in FSID 1250, EBD information for FSID 1250 / VLAN 150 is not advertised across that adjacency. Likewise, state belonging to an unpermitted FSID is not advertised. Control-plane reachability must never exceed data-plane reachability.

When an FSID/VLAN combination is removed from a Fabric Uplink, only the associated EBD and reachability state is withdrawn. The entire Fabric adjacency remains operational unless the underlying fabric relationship itself fails.

14. FSW Reachability Table

Each switch maintains an FSW Reachability Table. The EBD answers where an endpoint originates. The FSW Reachability Table references the candidate path set and installed next-hop group used to reach that Origin FSW-ID for the applicable FSID/VLAN forwarding context.

EBD:
FSID1250 / VLAN120 / MAC-X → FSW-C

FSW Reachability:
FSW-C / FSID1250 / VLAN120
  Candidate Path 1: A-B-C → Cost 20
  Candidate Path 2: A-D-C → Cost 20
  Candidate Path 3: A-E-C → Cost 30
  Installed Next-Hop Group: B, D

The control plane retains complete candidate-path awareness, including cumulative path cost and FSID/VLAN eligibility. The ASIC receives the selected local next-hop group for line-rate forwarding; the next-hop group is therefore the result of path computation, not a substitute for full-path knowledge.

15. Scalable Fabric Topology and Path-State Architecture

The enhanced fabric is intended to scale to datacenter environments containing thousands of Fabric Switches without multiplying endpoint state by every available path. The architecture therefore separates endpoint identity, shared topology state, candidate path computation, and installed forwarding state.

The physical Fabric topology is stored once as a link-state graph. Each link record contains the neighboring FSW identities, operational state, bandwidth-derived cost, and the FSID/VLAN combinations permitted across that link. A switch does not maintain a completely separate topology database for every VLAN.

For each destination FSW, the local switch derives complete candidate paths from that shared topology. Path eligibility is evaluated separately for the applicable FSID/VLAN forwarding context because an intermediate Fabric Uplink may permit one VLAN while pruning another.

Example from FSW-A to FSW-C:

FSID 1250 / VLAN 120
  A-B-C   Cost 20   VALID
  A-D-C   Cost 20   INVALID — VLAN 120 pruned on D-C
  A-E-C   Cost 30   VALID

FSID 1250 / VLAN 130
  A-B-C   Cost 20   VALID
  A-D-C   Cost 20   VALID
  A-E-C   Cost 30   VALID

This full-path awareness is required for loop prevention, cost comparison, failure convergence, FSID/VLAN pruning, and path-aware Fabric Congestion Notification. A switch cannot safely delegate those decisions to the next hop without knowing which complete paths are valid.

The EBD does not duplicate those path lists for every endpoint. An endpoint entry references its Origin FSW-ID. The Origin FSW reachability object then references the candidate-path set and the reusable local next-hop group selected from that set. Thousands of endpoints behind the same destination FSW can therefore share the same path and next-hop objects.

EBD:
  FSID/VLAN/MAC → Origin FSW-C
                    ↓
FSW Reachability:
  FSW-C / FSID/VLAN → Candidate Path Set
                         ↓
                    Next-Hop Group
                    ├─ FSW-B
                    └─ FSW-D

If a link fails or FCN temporarily changes path preference, the shared reachability or next-hop object is updated rather than rewriting every EBD entry associated with the destination FSW. Where several FSID/VLAN contexts resolve to an identical eligible path set, implementations may reuse the same path or next-hop object rather than storing duplicate copies.

To prevent an explosion of mathematically possible routes in dense topologies, implementations should retain all equal-best valid paths plus a bounded number of policy-approved alternate paths rather than every possible loop-free path. Exact path-set, FSW, endpoint, and hardware-table limits remain implementation dependent and require large-scale simulation and platform validation.

Topology distribution should reuse established link-state principles rather than inventing a new path-computation model. Fabric Switches automatically exchange authenticated neighbor and link-state information, sequence and age updates, synchronize topology state, and run shortest/equal-cost path computation internally. These mechanisms may reuse proven IS-IS/SPB concepts, but no separate operator configuration of IS-IS, NET addresses, areas, or another routing protocol is required.

Operator-facing configuration:

fabric-domain AI-FABRIC
  authentication key-id 1
  authentication password ********
  vedi enable

interface Ethernet1/49
  switchport mode fabric

Fabric discovery, authentication, FSW-ID exchange, link-cost derivation, FSID/VLAN capability exchange, topology synchronization, shortest-path computation, and forwarding installation are automatic consequences of Fabric Domain and Fabric Uplink configuration. Optional manual fabric-cost override may be provided for exceptional engineering requirements.

Operational commands should expose fabric-native abstractions such as show fabric topology, show fabric paths, show fabric reachability, and show fabric-segment rather than requiring routing-protocol-specific operational knowledge. The design principle is to reuse mature link-state mechanisms internally while hiding unnecessary protocol configuration complexity from the operator.

16. Layer-2 Path Cost

The enhanced fabric uses a bandwidth-derived Layer-2 cost model based conceptually on familiar spanning-tree path-cost behavior. This retains operational familiarity without retaining spanning-tree topology construction.

Received path cost: 20
Local link cost:    10
New advertised cost: 30

Lowest-cost valid paths are preferred. Equal-cost valid paths may remain concurrently active. Manual administrative cost override remains available. The architecture keeps the metric concept while discarding root-bridge election, designated-port topology, and redundant-path blocking.

17. Fabric Congestion Notification (FCN)

Multipath forwarding provides an opportunity to react to congestion before a fabric reaches severe queue buildup, packet loss, or repeated Priority Flow Control events. The architecture therefore includes an optional Fabric Congestion Notification (FCN) mechanism for short-lived, switch-to-switch congestion signaling between authenticated Fabric Switch neighbors.

FCN complements rather than replaces existing Ethernet flow-control mechanisms. IEEE 802.1Qbb Priority Flow Control may continue to provide local per-priority backpressure where lossless behavior is required, while FCN allows an upstream Fabric Switch to learn that a particular downstream path is becoming congested and temporarily reduce the amount of new traffic selected for that path.

A downstream Fabric Switch that detects persistent congestion may notify the relevant upstream neighbor using information such as the originating FSW-ID, affected destination FSW-ID or next-hop path, traffic class or priority, optional FSID context, congestion level, sequence number, and a short validity lifetime. Notifications are path-aware and are not flooded indiscriminately across the Fabric Domain.

Example:

FSW-A -> FSW-B -> FSW-D -> FSW-F
                     [congestion on link D-F]

FSW-D -> FCN -> FSW-B
FSW-B temporarily reduces selection of the path through FSW-D and increases use of another valid path toward FSW-F.

Congestion state is treated as a temporary forwarding attribute rather than as a permanent topology metric. Bandwidth-derived Layer-2 path cost remains the stable basis for reachability and shortest-path selection, while FCN may temporarily adjust multipath weighting among otherwise valid paths. A congestion event therefore does not require rewriting the underlying link cost or endpoint ownership state.

FCN should support several congestion levels or an equivalent compact congestion value rather than a simple congested/not-congested flag. Generation should use queue thresholds, persistence, rate-of-change, hysteresis, and dampening so that normal microbursts do not create excessive control-plane signaling. Congestion-clear information or short expiration timers allow upstream switches to restore normal weighting gradually after the condition subsides.

Immediate congestion response remains local to the Fabric Switch control plane and must not depend on an AI controller. The AI Control/Intelligence layer may separately consume aggregated FCN and queue telemetry to identify recurring congestion patterns, recommend longer-term path or capacity changes, and correlate congestion across the broader infrastructure.

The intended control model therefore has two loops: a fast switch-to-switch FCN loop for immediate adaptive multipath behavior, and a slower intelligence/telemetry loop for historical analysis and optimization.

18. Endpoint and Fabric Path Loop Prevention

VEDI endpoint provenance prevents switches from confusing local endpoints with remotely advertised endpoints. A locally validated endpoint always takes precedence over remote reachability claims unless the endpoint is part of an explicitly authorized multihoming relationship.

EBD advertisements may additionally carry an FSW path history. If a switch sees its own FSW-ID already present in the advertisement path, it rejects the update. Fabric topology computation uses link-state information and shortest-path calculation rather than spanning-tree blocking.

19. MAC Aging

Traditional inactivity-based MAC aging is no longer the primary endpoint-removal mechanism. A silent endpoint does not become unknown merely because it has not transmitted recently.

VEDI endpoint state changes primarily because of physical port failure, LACP state change, explicit endpoint withdrawal, endpoint mobility, validation failure, Origin FSW failure, or other authoritative local events. A longer validity or safety lifetime may still exist to prevent indefinitely stale records.

20. Layer-3 and Layer-2 VEDI Addressing

20.1 Layer-3 FSW

Where an SVI already exists, VEDI automatically reuses the configured SVI address. No additional VEDI IP is required.

interface Vlan120
  ip address 10.120.0.1/24
  vedi

20.2 Layer-2 FSW

A pure Layer-2 switch participates fully in VEDI/EBD without requiring an IP address. However, when active IP-level validation is desired, an optional per-VLAN VEDI IP may be configured.

fabric-segment 1250
  vlan 120
    vedi ip 10.120.0.2/24

This address exists for endpoint validation and does not convert the switch into a Layer-3 gateway.

21. ARP and Neighbor Discovery Suppression

When an endpoint sends an ARP request, the ingress FSW examines the validated EBD. If it has an authoritative active binding, the FSW replies locally and does not flood the ARP request through the fabric. IPv6 Neighbor Discovery may be handled similarly using validated EBD information.

VEDI responds only when the binding is authoritative and reachable. Entries in HOLD-DOWN, WITHDRAWN, UNREACHABLE, or SUPPRESSED states must not be used to provide false reachability.

22. Unknown Unicast

Unknown-unicast flooding should be disabled by default in a strict enhanced-fabric deployment. An unknown destination should trigger controlled endpoint resolution rather than unrestricted flooding. Resolution attempts must be rate-limited to prevent abuse.

23. Broadcast Handling

Generic broadcast traffic that cannot be safely suppressed is distributed only within the relevant FSID/VLAN domain. A short-lived fabric broadcast identity may include Origin FSW-ID, FSID, VLAN, and Broadcast Sequence. The ingress FSW generates this identifier and intermediate switches retain a short-lived duplicate-suppression cache.

If a broadcast returns to an FSW whose local EBD says the source MAC belongs to one of its own ports or logical LAG attachments, the returning copy is discarded.

24. Multicast

IGMP and MLD remain the endpoint-facing membership protocols. VEDI correlates group membership with validated endpoint information inside the relevant FSID/VLAN context. Edge FSWs may maintain group-to-local-port state, while transit FSWs need only maintain group-to-interested-downstream-FSW state. This avoids maintaining unnecessary per-endpoint multicast state throughout the fabric and prevents membership in one FSID from leaking into another.

25. LAG, vPC, and MLAG Operation

Existing LACP/LAG behavior remains unchanged. VEDI recognizes that an endpoint attached through a LAG belongs to a logical aggregated attachment rather than an arbitrary physical port.

Logical Attachment: LPG-1
Members:
  FSW-A / Port-Channel20
  FSW-B / Port-Channel20

Both switches retain their individual physical FSW identities. The same endpoint appearing on unrelated switches is considered a duplicate conflict unless explicitly authorized as multihomed.

26. Split-Brain Operation

When paired switches lose direct communication, Pair State transitions to DEGRADED. Both switches immediately advertise the degraded condition to the rest of the fabric. They retain their individual FSW-IDs, local endpoint state, synchronized endpoint provenance, and independent data-plane eligibility.

If both physical members can still reach a dual-homed endpoint, both may remain available for unicast load balancing. If one member loses local endpoint reachability, only that member is removed from the forwarding set. Pair-control failure therefore does not automatically become a data-plane outage.

27. EBD Hold-Down and Partition Recovery

When an Origin FSW becomes unreachable, its EBD entries are not immediately deleted. They enter HOLD-DOWN and are removed from active forwarding. If the FSW returns before the hold timer expires, incremental reconciliation occurs against retained state. If the timer expires, retained entries are removed and a later returning FSW performs fresh synchronization.

28. Endpoint Dampening

Flapping endpoint interfaces must not generate uncontrolled EBD churn. VEDI uses states such as STABLE, PENDING, and SUPPRESSED. Repeated instability accumulates a penalty and extends the suppression interval. Local forwarding responds immediately to physical failures, while fabric-wide EBD advertisements are consolidated.

29. Fabric-Link Dampening

Repeatedly flapping fabric links also require dampening. One exception is critical: loss of the last valid path to an FSW must be propagated immediately and must never be delayed by dampening. Alternative-path instability may be suppressed without hiding total loss of reachability.

30. BPDU Handling

STP-family protocols do not operate inside the enhanced fabric. If an active Fabric Uplink receives an STP/RSTP/MSTP BPDU, the port immediately enters BPDU-QUARANTINE. The switch blocks forwarding, removes the fabric adjacency, withdraws the path from FSW reachability, and raises a critical event.

Endpoint-facing access ports and 802.1Q trunks do not propagate BPDUs into the enhanced fabric. BPDUs received from such endpoints are locally discarded or terminated by default, while other permitted Layer-2 traffic continues to operate normally. The design does not depend on endpoints being BPDU-silent; an accidental or malicious endpoint BPDU must remain confined to the local edge port.

31. Standard 802.1Q Endpoint Interfaces

Virtualization hosts, wireless access points, wireless controllers, servers using VLAN subinterfaces, firewalls, storage appliances, and other VLAN-aware endpoint devices are treated as standard 802.1Q endpoint interfaces rather than requiring special fabric-specific port types. In the initial architecture, each endpoint-facing physical or logical interface is assigned to exactly one FSID.

interface Ethernet1/20
  switchport mode trunk
  switchport trunk allowed vlan 10,20,30
  fabric-segment 1250

VEDI may learn multiple MAC/IP bindings behind the trunk, keyed by FSID + VLAN + MAC. The switch does not need to know whether the downstream device is a hypervisor, wireless AP, controller, server, or another appliance. Existing 802.1Q semantics remain authoritative at the endpoint edge.

This improves scale and future compatibility because the fabric distinguishes only among ordinary endpoint access, endpoint trunking, and Fabric Uplinks, rather than maintaining device-type-specific edge modes.

32. Existing Layer-2 Compatibility

A major design requirement is that existing useful Layer-2 technologies continue operating normally wherever practical.

  • Ethernet framing;
  • IEEE 802.1Q VLAN tagging and trunking;
  • IEEE 802.1ad Q-in-Q on Fabric Uplinks;
  • IEEE 802.1Qbb Priority Flow Control where lossless per-priority backpressure is required;
  • LAG and LACP;
  • vPC/MLAG;
  • IGMP snooping;
  • MLD snooping;
  • storm control;
  • port security;
  • DHCP snooping;
  • ARP inspection;
  • server and appliance VLAN trunking;
  • wireless VLAN trunking;
  • virtualization switching.

The primary exception is STP/RSTP/MSTP root-bridge election and topology blocking, which are not used inside the enhanced fabric.

33. Greenfield Deployment Model

The architecture is intended initially for greenfield deployment. Legacy environments remain separated by Layer 3.

Enhanced VEDI/EBD Fabric
          |
        Layer 3
          |
Legacy STP/RSTP/MSTP Domain

A single Layer-2 domain should not mix spanning-tree-based forwarding and enhanced-fabric forwarding.

34. Failure Behavior

The architecture follows several fundamental failure principles:

  • Known-safe forwarding may temporarily survive control-plane degradation.
  • New endpoint origination may be restricted during control degradation.
  • Control-plane failure must never trigger uncontrolled flood-and-learn fallback.
  • Resource exhaustion must be explicit rather than silently changing forwarding behavior.

35. Benefits

Compared with conventional spanning-tree Layer 2, the proposed architecture may provide:

  • active use of redundant Layer-2 paths;
  • elimination of root-bridge dependency;
  • endpoint provenance;
  • validated FSID/MAC/IP/VLAN bindings;
  • reduced ARP/ND flooding;
  • reduced unknown-unicast flooding;
  • persistent endpoint awareness without ordinary MAC aging;
  • FSID/VLAN-aware distributed endpoint state;
  • reuse of overlapping endpoint VLAN numbers across independent FSIDs;
  • up to approximately 4094 directly identifiable Fabric Segments in the first-generation Q-in-Q transport, with a future path to a native 24-bit FSID;
  • independent topology and endpoint convergence;
  • multipath forwarding;
  • optional path-aware congestion notification and adaptive multipath weighting before severe congestion or repeated PFC activation;
  • improved duplicate-MAC handling;
  • explicit multihoming;
  • graceful split-brain behavior;
  • support for standard endpoint 802.1Q trunks without special device-type modes;
  • selective incremental control-plane synchronization;
  • scalable separation of EBD endpoint state from shared topology and candidate-path state;
  • FSID/VLAN-aware path eligibility without duplicating the physical topology database per VLAN;
  • automatic link-state topology operation without separate IS-IS configuration;
  • improved failure-domain visibility.

36. Relationship to Existing Technologies

The architecture should not be presented as replacing or rediscovering Shortest Path Bridging, EVPN, TRILL, or other previous work.

  • SPB demonstrates that Layer-2 shortest-path forwarding can replace spanning-tree blocking and that mature link-state mechanisms such as IS-IS can be reused for Layer-2 topology distribution and SPF computation.
  • TRILL demonstrates link-state-driven Layer-2 forwarding and additional loop protection.
  • EVPN demonstrates the value of explicit MAC/IP reachability distribution and multihoming control.
  • LACP/MLAG demonstrate active multi-link endpoint attachment.
  • IGMP/MLD provide established multicast membership mechanisms.
  • IEEE 802.1ad Q-in-Q provides an established two-level VLAN-tagging mechanism that can carry the initial FSID/VLAN namespace without defining a new first-generation wire format.
  • IEEE 802.1Qbb Priority Flow Control provides established local per-priority backpressure but does not itself provide path-aware upstream multipath avoidance.
  • IEEE 802.1Qau Quantized Congestion Notification provides useful precedent for explicit congestion feedback; FCN adapts the principle toward authenticated switch-to-switch, path-aware signaling among Fabric Switches rather than relying on an endpoint Reaction Point model.
  • IEEE 802.1ah Provider Backbone Bridging and SPBM provide additional precedent for separating service identity from ordinary customer VLAN identity and inform future scaling analysis.

The proposed contribution is the combination of these proven principles with a VEDI/EBD endpoint validation and provenance architecture intended to retain native Layer-2 operation without requiring a routed underlay and overlay merely to achieve endpoint-aware multipath Ethernet forwarding.

37. Areas for Further Development

  1. formal control-protocol message definitions;
  2. exact Fabric Domain handshake state machine;
  3. EBD journal format and synchronization protocol;
  4. exact automated Fabric link-state message format and state machine;
  5. validation of SPB/IS-IS link-state mechanisms that can be reused internally without exposing separate routing-protocol configuration;
  6. ASIC forwarding-table feasibility;
  7. maximum FSW, endpoint, candidate-path, and FSID/VLAN-context scaling;
  8. convergence simulation;
  9. broadcast duplicate-suppression design;
  10. multicast optimization;
  11. exact VEDI validation mechanisms for IPv4 and IPv6;
  12. security testing;
  13. failure and partition simulation;
  14. protocol-version interoperability;
  15. laboratory prototype implementation;
  16. comparison against STP/RSTP/MSTP, SPB, and EVPN-based fabrics.
  17. Q-in-Q ASIC scale, S-VID resource limits, and automatic MTU behavior across representative switching platforms;
  18. future native Layer-2 header feasibility with a 24-bit FSID while preserving the same FSID/VLAN/EBD semantics.
  19. FCN message format, queue-threshold/hysteresis behavior, adaptive multipath weighting, and interoperability with 802.1Qbb/PFC and existing congestion-control mechanisms.
  20. large-scale simulation of shared topology storage, per-link FSID/VLAN eligibility, bounded candidate-path sets, and reusable next-hop/path objects across thousands of Fabric Switches;
  21. convergence and control-plane load testing for automated link-state synchronization without explicit operator routing-protocol configuration.

38. Conclusion

Ethernet itself does not inherently require redundant links to remain blocked. Technologies such as Shortest Path Bridging have already demonstrated that a Layer-2 topology can use multiple physical paths when a topology-aware control plane is available.

The VEDI/EBD Enhanced Layer-2 Fabric architecture extends that principle by making endpoint identity and attachment authoritative rather than relying primarily on passive MAC learning. The proposed model combines validated endpoint intelligence, distributed endpoint provenance, unique FSW identities, FSID/VLAN-aware EBD propagation, Q-in-Q fabric transport, shortest-path Layer-2 reachability, active multipath forwarding, incremental synchronization, controlled broadcast and multicast behavior, multihoming and split-brain awareness, and preservation of established Ethernet technologies. Endpoint bindings remain separate from shared topology and path state, allowing complete FSID/VLAN-aware path knowledge to be maintained once per destination/path context rather than repeated for every MAC address.

The architecture does not attempt to replace Ethernet. Its first implementation deliberately reuses 802.1Q at endpoint-facing interfaces and 802.1ad Q-in-Q between Fabric Switches so that development can remain focused on replacing spanning-tree topology control. The direct S-VID-to-FSID mapping keeps the initial design simple to operate and troubleshoot, while a future native 24-bit FSID may be introduced only if greater Fabric Segment scale becomes necessary. Its topology control plane reuses mature link-state concepts internally but is intended to self-form from Fabric Domain and Fabric Uplink configuration without requiring a separately configured routing protocol.

The design remains a conceptual technology for further development and should be validated through detailed protocol engineering, simulation, standards analysis, hardware feasibility testing, and prototype implementation before production adoption.

© 2026 Christopher Soans. All rights reserved.

This work is licensed under a Creative Commons Attribution 4.0 International License (CC BY 4.0).