marking and queuing on Nokia IP routers
Scope
Nokia IP routers (SR OS on 7750 SR/7950 XRS) implement queuing with a fixed set of forwarding classes, a scheduler policy per port or per SAP, a slope policy for WRED, and a port scheduler for port-level bandwidth. This domain tests that architecture and the marking that feeds it.
Core rules
- There are eight forwarding classes: be, l2, af, l1, h1, h2, ef and nc. Each maps to a queue; the FC-to-queue mapping is configured in the SAP ingress/egress or network QoS policy.
- A scheduler policy defines a hierarchy: scheduler-tiers (up to four levels) for H-QoS, child schedulers, and a root. Each scheduler has CIR, PIR, priority (1–8, 1 highest) and weight.
- A slope policy defines high-slope and low-slope WRED parameters: start-average, max-average and max-probability, per queue.
- The port scheduler allocates bandwidth across the port's queues and scheduler hierarchy; it is the final arbiter on egress.
- H-QoS uses scheduler-tiers to shape each subscriber or service before the port scheduler, preventing one subscriber from consuming another's bandwidth.
- Network QoS policies classify and mark on network interfaces; SAP QoS policies do the same on service access points.
Worked example
A 10G port carries three subscribers, each with a 1G H-QoS shaper and a voice queue. Voice is FC ef with priority 1; data is af with priority 4. The port scheduler sees the sum of the subscriber shapers, not raw line rate. If subscriber A bursts, its tier-1 shaper drops or delays its own excess before the port scheduler's child arbitration, so B and C keep their CIR. Inside each subscriber, SP for ef guarantees voice latency.
Traps
- Thinking FC numbers are queue numbers; they are separate and mapped by policy.
- Using priority 8 as highest - in SR OS priority 1 is highest.
- Forgetting that a slope policy must be attached to the queue/policy to have any effect.
- Confusing port scheduler bandwidth with the physical port speed when an aggregate shaper is configured.