Network QoS

1. Network QoS

2. End-to-end QoS • The end-to-end QoS is dependent on the QoS provisioned by each of the domains • Requirements of end-to-end QoS: • Per-hop QoS • Routing and traffic engineering • Signaling and provisioning

3. Per-Hop Behavior • Each router is the smallest controllable convergence and divergence point for numerous unrelated flows of packets • Routers have buffer to handle the transient high rate of inflow of packets • End to end latency = transmission delay + processing delay at the routers • Delay introduced due to the router’s congestion-induced buffering is not predictable • Packets are lost when the router buffers are full

4. Classification, Queuing and Scheduling • QoS metrics: latency, jitter, packet loss • Issues: QoS characteristics of links, dynamics of queue utilization, and queue management within each router • Can be done through the CQS model: • Classification: arriving packets are classified • Queuing: multiple queues with different queue management and packet discard policies • Scheduling: mechanism to interleave packets from each queue

5. Edge-to-Edge Behavior • End-to-end service is formed from both edge-to-edge and per-hop behaviors • Network designers face a trade-off between the number of traffic classes carried by the networks and the number of traffic classes their router’s CQS architectures can handle • The existing destination-based shortest path routing algorithms in the Internet are not necessarily optimal for different classes of traffic across an arbitrary mesh of routers and links

6. Edge and Core? • The edge-and-core model allows core routers to leverage hardware implementations while leaving complex processing to edge routers • Edge routers can classify large number of traffic classes and the core routers will facilitate a limited number of traffic classes and could possibly employ behavioral aggregation • Edge router should smooth out the burstiness of traffic entering the network

7. Traffic Shaping • The focus of CQS architecture is the isolation of traffic in each queue from the burstiness in another queue. • Managing customer expectation through rate capping is essential • Placing an upper bound on the maximum bandwidth (on minimum inter-packet interval) available to a traffic class is known as “traffic shaping” • A shaping scheduler is configured to provide both a minimum service interval and a maximum service interval • A simple shaping scheduler: “leaky bucket”

8. Policing • Policing involves dropping of packets when too many packets arrive in too short an interval • Can be implemented through counters for each traffic class • Dropping of packets translate to the notification of congestion to the TCP sources • Policing protects the network by continuing to drop packets that exceed the allowable parameters

9. Marking and Reordering • Policing can be softened by marking the packets that exceed certain threshold • The marked packets would be treated as low priority compared to the unmarked packets • The queue management algorithms can begin dropping the marked packets before beginning to drop the unmarked ones • Marking improves the bandwidth utilization, but care should be taken to avoid packet reordering in the implementation.