Dynamic VLAN Demultiplexing

Automatic subscriber VLAN interface creation — no pre-provisioning, no static configuration, no manual intervention.

The Problem

In traditional BNG deployments, every subscriber VLAN must be pre-configured on the BNG before a subscriber can connect. For networks with thousands of VLANs — especially QinQ deployments with S-Tag/C-Tag combinations — this means:

• Thousands of lines of static VLAN configuration
• Manual provisioning for every new subscriber
• Configuration drift when subscribers move or churn
• Stale interfaces consuming resources for inactive subscribers

How AthenaBNG Solves It

The abng-demux daemon listens on trunk interfaces using AF_PACKET sockets with BPF filters. When it sees a PPPoE Discovery (PADI) or DHCP Discover frame on a VLAN that doesn't yet have an interface, it automatically creates the entire VLAN interface chain.

How It Works

1. Capture — BPF filter on the trunk captures only PPPoE Discovery and DHCP Discover frames (not data traffic)
2. Parse — Extract VLAN tag stack from the Ethernet frame (supports arbitrary nesting)
3. Validate — Check against rate limits, deduplication window, and max interface count
4. Create — Build VLAN interface chain via netlink (outer → inner)
5. Notify — Tell abngd the new interface is ready
6. Bind — abngd tells accel-ppp or abng-dhcpd to listen on the new interface

VLAN Support

Type	Ethertype	Example
802.1Q	0x8100	trunk0.100
802.1ad (QinQ)	0x88A8	trunk0.200.100
Triple-tagged	Mixed	trunk0.111.200.100

Interface naming follows the VLAN stack: {trunk}.{outer}.{stag}.{ctag}

Safety Features

• Rate limiting — Max interface creations per second (default: 100)
• Deduplication — Requires multiple matching packets within a time window before creating
• Hard limit — Maximum total dynamic interfaces (prevents resource exhaustion)
• Idle cleanup — Background task removes interfaces with no active sessions and no traffic

Performance

abng-demux is built for speed:

• BPF filtering at the socket level — only control-plane frames reach userspace
• Pre-allocated buffers — no per-packet memory allocation
• Kernel VLAN tag recovery — handles hardware VLAN offload transparently via PACKET_AUXDATA
• O(1) interface lookups — HashMap-based state tracking

Configuration

demux:
  enabled: true
  trunks:
    - interface: eth1
      vlan_ranges:
        - start: 100
          end: 4094
  idle_timeout: 300
  max_interfaces: 8192
  creation_rate_limit: 100

Why This Matters

For an ISP with 2,000 subscribers across 200 S-Tags and 10 C-Tags per S-Tag, that's potentially 2,000 VLAN interfaces. With AthenaBNG, you configure zero of them. They appear when subscribers connect and disappear when they leave.