Introduction

Three modes of ingress SAP queuing are supported for multipoint services (IES, VPLS and VPRN); service, shared, and multipoint shared. The same ingress queuing options are available for IES and VPLS subscriber SLA profile instance queuing.

Normal or service queuing is the default mode of operation for SAP ingress queuing. Service queuing preserves ingress forwarding bandwidth by allowing a service queue defined in an ingress SAP QoS policy to be represented by a group of hardware queues. A hardware queue is created for each switch fabric destination to which the logical service queue must forward packets. For a VPLS SAP with two ingress unicast service queues, two hardware queues are used for each destination forwarding engine the VPLS SAP is forwarding to. If three switch fabric destinations are involved, six queues are allocated (two unicast service queues multiplied by three destination forwarding complexes equals six hardware queues). Unicast service queue mapping to multiple destination based hardware queues demonstrates unicast hardware queue expansion. Service multipoint queues in the ingress SAP QoS policy are not expanded to multiple hardware queues, each service multipoint queue defined on the SAP equates to a single hardware queue to the switch fabric.

When multiple hardware queues represent a single logical service queue, the system automatically monitors the offered load and forwarding rate of each hardware queue. Based on the monitored state of each hardware queue, the system imposes an individual CIR and PIR rate for each queue that provides an overall aggregate CIR and PIR reflective of what is provisioned on the service queue.

To avoid the hardware queue expansion issues associated with normal service based queuing, the system allows an ingress logical service queue to map to a single hardware queue when shared queuing is enabled. Shared queuing uses two passes through the ingress forwarding plane to separate ingress per service queuing from the destination switch fabric queuing. In the case of shared queuing, ingress unicast service queues are created one-for-one relative to hardware queues. Each hardware queue representing a service queue is mapped to a special destination in the traffic manager that ‛forwards’ the packet back to the ingress forwarding plane allowing a second pass through the traffic manager. In the second pass, the packet is placed into a ‛shared’ queue for the destination forwarding plane. The shared queues are used by all services configured for shared queuing.

When the first SAP or SLA profile instance is configured for shared queuing on an ingress forwarding plane, the system allocates eight hardware queues per available destination forwarding plane, one queue per forwarding class. Twenty four hardware queues are also allocated for multipoint shared traffic. The shared queue parameters that define the relative operation of the forwarding class queues are derived from the Shared Queue policy defined in the QoS CLI node. Unicast service queuing with shared queuing enabled demonstrates shared unicast queuing. SAP or SLA profile instance multipoint queuing is not affected by enabling shared queuing. Multipoint queues are still created as defined in the ingress SAP QoS policy and ingress multipoint packets only traverse the ingress forwarding plane a single time, as demonstrated in Multipoint queue behavior with shared queuing enabled.

Enabling shared queuing may affect ingress performance because of double packet processing through the service and shared queues.

Ingress multipoint shared queuing is a variation to the unicast shared queuing defined in Ingress shared queuing. Ingress unicast service queues are mapped one-for-one with hardware queues and unicast packets traverse the ingress forwarding plane twice. In addition to the above, the multipoint queues defined in the ingress SAP QoS policy are not created. Instead, multipoint packets (broadcast, multicast and unknown unicast destined) are treated to the same dual pass ingress forwarding plane processing as unicast packets. In the first pass, the forwarding plane uses the unicast queue mappings for each forwarding plane. The second pass uses the multipoint shared queues to forward the packet to the switch fabric for special replication to all egress forwarding planes that need to process the packet.

The benefit of defining multipoint shared queuing is the savings of the multipoint queues per service. By using the unicast queues in the first pass and then the aggregate shared queues in the second pass, per service multipoint queues are not required. The predominate scenario where multipoint shared queuing may be required is with subscriber managed QoS environments using a subscriber per SAP model. Usually, ingress multipoint traffic is minimal per subscriber and the extra multipoint queues for each subscriber reduces the overall subscriber density on the ingress forwarding plane. Multipoint shared queuing eliminates the multipoint queues sparing hardware queues for better subscriber density. Multipoint shared queuing using first pass unicast queues demonstrates multipoint shared queuing.

One disadvantage of enabling multipoint shared queuing is that multipoint packets are no longer managed per service (although the unicast forwarding queues may provide limited benefit in this area). Multipoint packets in a multipoint service (VPLS, IES and VPRN) use significant resources in the system, consuming ingress forwarding plane multicast bandwidth and egress replication bandwidth. Usually, the per service unicast forwarding queues are not rate limited to a degree that allows adequate management of multipoint packets traversing them when multipoint shared queuing is enabled. It is possible to minimize the amount of aggregate multipoint bandwidth by setting restrictions on the multipoint queue parameters in the QoS node’s shared queue policy. Aggregate multipoint traffic can be managed per forwarding class for each of the three forwarding types (broadcast, multicast or unknown unicast – broadcast and unknown unicast are only used by VPLS).

A second disadvantage to multipoint shared queuing is the fact that multipoint traffic now consumes double the ingress forwarding plane bandwidth because of dual pass ingress processing.