Scaling Subscribers on the TFA ASIC with QoS

The TFA ASIC on the ES2 10G LM supports a total of 32,000 nodes; however, it requires that each queue stack above a node at both level 1 and level 2, and it cannot skip a level in the scheduler hierarchy. The FFA ASIC also requires that each queue stack above a node at both level 1 and 2, but it also offers more nodes, so the scheduler hierarchy requirement is not as visible. The EFA ASIC does not require queues to stack above any level.

Because the TFA ASIC cannot skip a level in the hierarchy and also offers a smaller amount of nodes, scaling subscribers for triple-play configurations can exhaust node resources. For example, the ethernet-default QoS profile specifies both an IP and a VLAN node. Configuring 16,000 IP over VLAN subinterfaces consumes all 32,000 nodes, with no node resources remaining for other traffic-class groups. By carefully configuring queues on the TFA ASIC, you can scale up to 16,000 subscribers for multiple traffic-class groups in a triple-play configuration.

To conserve nodes on the TFA ASIC, you could apply one of the following configurations:

In non-default traffic-class groups, you can configure a group node and VLAN queues. Although this apparently does not consume nodes, it does consume a hidden, phantom node for each queue, to satisfy the level requirement of the TFA ASIC.

Alternatively, use group nodes and shadow nodes.

We recommend that you configure an Ethernet shadow node in the group with the following QoS profile rule:

host1(config-qos-profile)#ethernet shadow-node group groupname

This rule stacks another node over the group node, so all VLAN queues are stacked above the single shadow node. No nodes are consumed in the traffic-class group.

Related Documentation