• Chapter 1: Cisco VOIP Implementations
• Chapter 2: IP Quality of Service
• Chapter 3: Classification, Marking, and NBAR
• Chapter 4: Congestion Management and Queuing
• Chapter 5: Congestion Avoidance, Policing, Shaping, and Link Efficieny Mechanisms
• Chapter 6: Implementing QoS Pre-Classify and Deploying End-to-End QoS
• Chapter 7: Implementing AutoQoS
• Chapter 8: Wireless LAN QoS Implementation
• Chapter 9: 802.1x and Configuring Encryption and Authentication
• Chapter 10: WLAN Management

Chapter 5: Congestion Avoidance, Policing, Shaping, and Link Efficieny Mechanisms

Congestion avoidance is implemented to avoid tail drop, which occurs when there is no room left in a queue for incoming packets.

Tail drop is not selective; less aggressive flows are not preferred to aggressive flows, thus no QoS can be provided.

TCP global synchronization occurs when tail dropping of packets forces flows to cycle between small and large windows.

TCP starvation occurs when stateless protocols like UDP fill available queue space before the throttled TCP flows.

Random Early Detection (RED)

When RED is implemented, packets are randomly dropped before the queue becomes full.

The rate of drop increases as the queues nears its maximum size.

RED mitigates the problem of TCP synchronization.

Configuration parameters:

• Minimum threshold - Below this no packets are dropped
• Maximum threshold - Above this all packets are dropped
• Mark Probability Denominator (MPD) - An integer specifying the base probability of drop

Weighted Random Early Detection (WRED)

WRED is RED with the added capability of favoring prioritized traffic, based on the IP precedence or DSCP.

Class-Based WRED (CBWRED)

CBWRED is WRED implemented inside a CBWFQ system.

CBWRED is applied to a CBWFQ class (under a policy map) with random-detect.

CBWRED operates on IP precedence by default but can be configured to evaluate DSCP.

Each precedence/DSCP value can be configured with a unique MPD and minimum and maximum thresholds.

Configuration example:

policy-map Foo
 class Precedence_Based_WRED
  bandwidth 100
  random-detect
 class DSCP_Based_WRED
  bandwidth 100
  random-detect dscp-based

Traffic Shaping and Policing

Policing

Policing restricts the amount of bandwidth consumed by traffic.

Traffic which exceeds the policed threshold can be dropped or remarked to a lower QoS.

Purposes:

• Enforcing subrate access; limiting available bandwidth to less than that of the physical interface
• To limit the traffic rate per class
• To remark traffic not conforming to an SLA

Policing can be applied inbound or outbound on an interface.

Shaping

Shaping buffers excess traffic for transmission, introducing a delay.

Purposes:

• To slow the rate at which traffic is sent to a congested destination
• To comply with a subscribed rate (bandwidth cap)
• To transmit traffic from different classes at different rates

Shaping can only be applied outbound.

Shaping introduces variable delay when traffic is buffered.

Shaping can be configured to respond to network conditions and signals, such as frame relay Backward Explicit Congestion Notifications (BECNs).

Link Efficiency Mechanisms

Most link efficiency mechanisms are only required or supported on slow links.

Layer 2 Payload Compression

Layer 2 payload compression is implemented on a link-by-link basis, and compresses the entire layer 2 payload.

Compression introduces a processing delay, but reduces serialization delay and increases available bandwidth.

Compression can be performed in hardware or software; compression performed in software is CPU-intensive and not recommended.

Header Compression

Header compression can be used with TCP or RTP. Only headers are compressed, not payload.

Like L2 payload compression, header compression is implemented on a link-by-link basis.

Link Fragmentation and Interleaving (LFI)

Large frames are fragmented and interleaved with smaller, high-priority frames to reduce jitter.