Edit

QoS

Traffic Shaping

Commited Burst

  • This is the amount of traffic (in bits) allowed to be sent in each Tc interval at the Committed Information Rate (CIR).

  • Bc = CIR × Tc / 1000

    • CIR is in bits per second (bps)

    • Tc is in milliseconds (ms)

    • Bc is in bits

  • Example: If CIR = 512,000 bps and Tc = 20 ms: Bc = 512,000 × 20 / 1000 = 10,240 bits -> This means the shaper allows 10,240 bits every 20 ms.

Excess Burst

  • This is additional traffic allowed beyond Bc if there’s unused bandwidth (i.e., tokens left in the bucket). It’s optional and used in three-color marking (conform, exceed, violate).

  • Usually Bc = Be, but can be adjusted based on how much burstiness you want to allow.

    • Be = Bc: moderate bursting

    • Be = 2 x Bc: aggressive bursting

  • This allows the router to send more than Bc if tokens have accumulated (i.e., if traffic was below CIR previously).

  • With Be = 0, excess traffic above Bc/Tc is immediately dropped or delayed, which can be fine-tuned if you have bursty traffic patterns.

Queue Limit

  • The number of packet in the buffer

  • Rule of thumb:

    • For WAN shaping (e.g. 10–500 Mbps): Queue limit of 10–50 ms worth of traffic

    • Can calculate that based on the interface speed and average packet size

  • Example: link speed = 500Mbps

    • Average packet size: 1000 bytes (8000 bits)

    • Desired buffering: 50 ms worth of traffic

    • Formula: Buffer size (in packets) = (CIR × buffering time) / average packet size

      • (500,000,000 × 0.05) / 8000 ≈ 3125 packets

  • Memory ≈ queue depth × average packet size

  • Check Memory Usage:

    • show platform hardware qfp active datapath utilization summary

    • show platform hardware qfp active statistics drop

    • show memory statistics history table

  • EEM Event to check interface drop

Bandwidth Delay Product

  • The BDP is a foundational concept in networking that represents the maximum amount of data “in flight” on a link.

  • It’s calculated as: BDP = Bandwidth × Round-Trip Time (RTT)

  • This helps determine:

    • How big queues should be to absorb bursts without dropping packets

    • How much buffering is needed to maintain throughput without inducing latency

  • Cisco’s Enterprise QoS Solution Reference Network Design (SRND) and Catalyst QoS Deep Dive sessions recommend: Queue size ≈ 10–50 ms worth of traffic, depending on:

    • Link speed

    • Application sensitivity (e.g., voice vs. bulk data)

    • Hardware buffer capacity

  • This is why the common formula is: Queue size (bits) = Bandwidth × 0.01 to 0.05 seconds. Then divide by average packet size to get the queue limit in packets.

  • Why It Matters

    • Too small a queue → drops during microbursts

    • Too large a queue → bufferbloat and latency

    • Just right → smooth traffic, minimal jitter, and efficient shaping

  • If tuning for real-time traffic (like VoIP), Cisco also recommends strict priority queues with burst bytes sized to 1.5× the BDP of the voice stream.

  • Example using 500 Mbps link and a target latency budget to size queues for different traffic types.

    • Let’s assume you have three main traffic types:

      • Voice (EF) — latency-sensitive

      • Video (AF41) — bursty but delay-tolerant

      • Bulk data (AF11/BE) — best-effort

    • Target queuing delay is:

      • Voice: 10 ms

      • Video: 50 ms

      • Bulk: 100 ms

      • Assume average packet size = 1000 bytes (8000 bits)

    • So

      • Voice Queue: BDP = 500,000,000 × 0.01 = 5,000,000 bits → 625 packets

      • Video Queue: BDP = 500,000,000 × 0.05 = 25,000,000 bits → 3125 packets

      • Bulk Queue: BDP = 500,000,000 × 0.1 = 50,000,000 bits → 6250 packets

    • Sample configuration

    Reference

PreviousCommandsNextReference

Last updated