Help us improve your experience.

Let us know what you think.

Do you have time for a two-minute survey?

 
ON THIS PAGE
 

Use Case and Reference Architecture

As discussed in the Solution Benefits section, the BBE JVD architecture is based on the DBAS concept. The following solution model shows Juniper DBAS architecture.

Figure 1: Juniper DBAS Solution Architecture A diagram of a cloud network Description automatically generated

In this architecture, the overlay network infrastructure has effectively become a simplified, data centre style access fabric using Multiprotocol Label Switching (MPLS) Segment Routing (SR) as transport underlay and EVPN as service overlay. And therefore, it can run on a variety of Juniper underlay platforms that support AGN “spine” functions in an access and aggregation capacity.

Figure 2: Juniper Metro DBAS Key Elements A diagram of a company Description automatically generated

Figure 2 shows key elements of the DBAS architecture. The BBE/BNG service leaves enable the broadband services to provide Point-to-Point Protocol over Ethernet (PPPoE) and IP over Ethernet (IPoE) sessions over the EVPN Pseudowire Headend Termination (PWHT). This architecture also provides access-agnostic connectivity for residential, retail/business, and/or wholesale services.

The AN leaves and BBE/BNG service leaves are horizontally scaled. and can aggregate and integrate PON access. Disaggregating larger centralized devices provides significant advantages, such as lower power consumption, reduced space requirements, and decreased cooling demands. This approach enables agile, on-demand scalability.

The transport underlay in the access fabric is based on SR MPLS, enabling operational simplicity and fast convergence with Loop Free Alternate Fast Re-Route (LFA FRR). In case of BBE JVD, rapid topology convergence is achieved with Topology Independent Loop Free Alternates (TI-LFA) fast reroute. In cases where the CGNAT service gateway is remote from the core network (see Juniper Scale-Out Stateful Firewall and Source NAT for Enterprise — JVD ), the BBE/BNG service leaves and the Access Fabric AGN spines implement BGP label unicast to enable segmentation and isolation between core and access fabric and facilitate seamless MPLS integration across the domains.

The service overlay for unicast services is based on EVPN Virtual Private Wire Service (VPWS), which enables per Access Node transport. It supports EVPN Flexible Cross Connect (FXC), which enables multiplexing of multiple access nodes in the same EVPN transport, improving operational simplicity in the access fabric.

The access fabric provides IP multicast transport for IP Television (IPTV) services. The EVPN transport enables the broadband access to be load balanced across the BNG service leaves cluster. By using a different designated forwarded priority, it provides active/standby connectivity to backup BBE/BNGs.

BBE/BNG service leaves enable:

  • PWHT for the EVPN service
  • PPPoE, IPoE, and IP multicast for retail unicast and multicast services
  • Link Access Concentrator (LAC) and MPLS VPN for wholesale services

Architecture Functional Layers

  • Metro Ethernet Network (MEN): The reference architecture deploys a spine-leaf access topology to provide the physical redundancy (wherever appropriate) for L2 to connect to a PWHT for BNG services:
  • The Transport Underlay layer includes:
    • SR MPLS using ISIS IPv4 (Metro and Core)
    • iBGP-LU
    • MP-BGP
  • The Services Overlay for PWHT EVPN E-LINE includes:
    • (EVPN-VPWS) multihomed without FXC
    • EVPN E-LINE (EVPN-VPWS) multihomed with FXC

BBE/BNG Solution

Juniper operates two redundancy models for DBAS using BBE/BNG solutions to achieve minimal subscriber traffic drop during failover and keep the revert time predictable (i.e., it is the same as failover):

Stateless Rapid Reconnect (RR) model provides the following benefits:

  • This model can be leveraged for PPPoE, DHCP C-VLAN, and static VLAN methods for relay and server.
  • Use this model to optimize upgrades, i.e., fast restoration of best-effort traffic, which can be applied on access interfaces such as PS, GE, XE, and AE.
  • There is no overhead in synchronising the subscriber state and client information between the BNGs.

A drawback of this model is:

  • There is no background programming of CoS, firewall, and services structure upon node failure.

For more information, see BNG Redundancy for DHCP Subscribers Using Packet Triggered Based Recovery .

Figure 3 shows Juniper BBE redundancy models. Juniper supports traditional stateful and stateless mechanisms of operation as well as new models with Stateless Rapid Reconnect and N:1 Stateful model with PFE oversubscription. This JVD is based on the new model of Stateless Rapid Reconnect.

Figure 3: Juniper BNG Redundancy Traditional Models A diagram of a cloud network Description automatically generated
Figure 4: Juniper BNG Redundancy New Models A diagram of a cloud computing system Description automatically generated

Figure 5 shows the operation of Stateful N:1 Redundancy model. This model shows the Active Lease subscriber synchronization between the Master and Backup BNG.

Figure 5: Juniper BNG Stateful N:1 Redundancy Model Juniper BNG Stateful N:1 Redundancy Model

Figure 6 shows Stateless Rapid Reconnect. There is no synchronization between the Primary and Backup BNG nodes. In case of the primary BNG node failure, the first data packet of an existing subscriber session is used to trigger the creation of a subscriber interface on the Backup BNG. After the expiry of the DHCP lease, a new DHCP process is used to create the subscriber sessions on Backup BNG.

Figure 6: Juniper BNG Stateless Rapid Reconnect Flow A diagram of a server Description automatically generated

Figure 7 presents the call flow procedures during BNG failover between subscriber device (CPE) to Primary and Backup BNG. The diagram explains how the session is recreated on the Backup BNG.

Figure 7: Juniper BNG Stateless Rapid Reconnect Process Juniper BNG Stateless Rapid Reconnect Process

Before the failover:

  • DHCP over Dynamic VLAN stack is present at Primary BNG.
  • There are no subscribers at the Backup BNG.

At the point of failover, the first data packet that arrives on the Backup BNG triggers the creation of a Dynamic VLAN and a Dynamic IP subscriber. The Dynamic IP subscriber stacks over the Dynamic VLAN.

If the primary BNG continues to show a failure state when the DHCP renew occurs, it is addressed as NAK, and it restarts the DORA process. This creates the DHCP subscriber at backup BNG and stacks DHCP over dynamic VLAN. The dynamic IP subscriber created earlier (at step 2 above) is deleted.

N:1 Stateful Model (N<4) (formerly known separately as 1:1 Stateful Model and M:N Stateful Model: In this model, Active Lease Query (ALQ) and Bulk Lease Query (BLQ) are used to synchronise subscribers from the primary BNG(s) to the backup BNG.

The benefits of this model are:

  • N+1 (N<4) PFE oversubscription using pseudowire (PS) interface (MPLS PWHT, EVPN VPWS).
  • One chassis acts as a backup for Nx BNGs (Juniper recommends four primary BNGs to one backup BNG as a ratio).
  • Upon failover, best-effort traffic forwarding immediately resumes. In the background, CoS, firewall, and services are programmed.

Drawbacks of this model Include:

  • DHCP subscribers only, therefore no PPPoE support.
  • Single failure design (one failover at a time; second failover needs revert of the first failover to complete)

For more information, see M:N Subscriber Redundancy on BNG .

Baseline Features

The baseline features required for this JVD include:

  • EVPN: EVPN E-LINE (EVPN-VPWS) with and without FXC
  • Routing: SR MPLS, ISIS, MP-BGP, iBGP-LU, eBGP, BFD, Route Reflection, IPv4
  • Switching: ESI LAG, VLAN (802.1q), VLAN QinQ (802.1ad)