Help us improve your experience.

Let us know what you think.

Do you have time for a two-minute survey?

 
 

Routing Protocols

  • BGP support for deterministic path forwarding (DPF) in a CLOS network (QFX5130-32CD, QFX5130-48C, QFX5130-48CM, QFX5700, QFX5700E, QFX5230-64CD, QFX5240-64OD, and QFX5240-64QD)—BGP Deterministic Path Forwarding (DPF) divides a physical fabric into multiple logical fabrics so that different flows can be mapped to different logical fabrics to serve the requirements of the flows. Single hop External BGP (EBGP) might not meet all DC flow requirements, especially for drop and latency sensitive AI ML flows. BGP DPF colors the single-hop EBGP session on each link with a fabric color. If a link belongs to the red fabric, the EBGP session over the link is colored red. A route with no color community is advertised over any colored or uncolored EBGP sessions.

    To configure fabric colors for BGP neighbors include the fabric-color com-name statement at the [edit protocols bgp] hierarchy level. You can also configure this feature at the BGP group or neighbor level.

    To advertise colored IPv4 or IPv6 routes, include the fabric-advertise statement at the [edit protocols bgp] hirearchy level.

    [See BGP Deterministic Path Forwarding in a CLOS Network .]

  • IS-IS multi-instance support over a single interface (ACX7020, ACX7024, ACX7100, ACX7332, ACX7348, ACX7509, PTX10001-36MR, PTX10002-36QDD, PTX10004, PTX10008, PTX10016, QFX5130, QFX5220, QFX5230-64CD, QFX5240, QFX5241-32OD, and QFX5700)—We have enhaced the IS-IS multi-instance feature to support multiple IS-IS instances on the same logical interface with instance identifier TLV 7.

    Include the instance-id statement at the [edit protocols isis-instance name hierarchy level.

    [See How to Configure Multiple Independent IGP Instances of IS-IS.]

  • GLB multi-link support on IP Fabric (QFX5240-64OD and QFX5240-64QD)—We are extending Global Load Balancing (GLB) to support multiple paths between spine and top-of-rack switches on a 3-stage Clos IP fabric.

    To enable GLB for multi-link on a 3-Clos IP fabric, include the glb-multilink-mode max-val|avg-val statement at the [edit forwarding-options enhanced-hash-key] hierarchy level. By default, the spine advertises the average quality of all links. Enable GLB globally at [edit protocols bgp] hierarchy level.

    [See Configure GLB on 3-CLOS IP Fabric with Multilinks.]

  • Path quality profile sharing for GLB multi-link support on IP Fabric (QFX5240-64OD, and QFX5240-64QD)—While we support larger Clos networks and more GPUs, the TH5 chipset can only support 64 profiles. In Clos networks with five or more stages, some nodes, like super spines exceed 64 next-next-hop nodes. We can reuse the profiles under specific conditions to support more than 64 next-next-hop nodes. We support GLB for Clos networks with profile sharing in hyperscaler artificial intelligence/machine learning (AI/ML) fabrics containing tens of thousands of leaves or GPUs.

    [See profile-sharing.]

  • Support for BGP bandwidth unequal load balancing for EVPN-VXLAN routes (QFX5130 and QFX5700)—We have extended the asymmetric traffic distribution support to EVPN-VXLAN routes. You can enable this feature based on path cost. This approach optimizes BGP traffic across paths with different bandwidths. The control plane calculates balance values for each path, and the Packet Forwarding Engine uses these values to distribute traffic proportionally ensuring efficient utilization of network resources.

    [See BGP Link-Bandwidth Community.]

  • Support for 256-way ECMP (QFX5130-32CD, QFX5230-64CD, QFX5240-64OD, and QFX5240-64QD)—Use this feature to increase the number of direct BGP peer connections, improve latency, and optimize data flow by configuring up to 256 ECMP next hops for external BGP peers.

    [See Example: Load Balancing BGP Traffic.]

  • Proxy ARP and NDP Proxy Support with VRRP (PTX10001-36MR, PTX10002-36QDD, PTX10004, PTX10008 and PTX10016)—Proxy ARP and NDP Proxy support with VRRP allows a router to use the VRRP virtual MAC address when responding to proxied ARP, NDP, or DAD requests. When an interface is configured with both proxy and VRRP, only the VRRP master responds to proxy requests, and the response uses the VRRP virtual MAC instead of the physical interface MAC. This behavior applies to restricted and unrestricted modes for Proxy ARP, NDP proxy, and DAD proxy. By using the virtual MAC address, the feature ensures consistent address resolution and seamless connectivity during VRRP failover events, preventing disruption when mastership changes occur. This feature is supported on Junos OS Evolved platforms.