Low Latency DOCSIS

This document describes how to configure Low Latency DOCSIS (LLD) on the Cisco cBR Series Converged Broadband Router.

Overview

Cablelabs introduced Low Latency DOCSIS technology (LLD) in DOCSIS 3.1 to further reduce latency for applications that are sensitive to delays. LLD technology uses separate service flows, Classic Service Flow and Low Latency Service Flow (LL SF), for traffic that may cause queue building (QB, for example, TCP CUBIC) and traffic that does not cause queue building (NQB, for example gaming, L4S).

Active Queue Management

Table 1. Feature History

Feature Name

Release Information

Feature Description

Support for Active Queue Management(AQM)

Cisco IOS XE Cupertino 17.9.1y

Active Queue Management (AQM) is one of the critical components of LLD. AQM allows routers to control the queue length and delay of packets.

Active Queue Management (AQM) is one of the critical components of Low-Latency DOCSIS (LLD). AQM allows routers to control the queue length and delay of packets.

Traditional Queue Management can lead to tail dropping when the queue is full. By dropping packets before buffer overflow, AQM allows routers to control the queue length and delay of packets.

Benefits of AQM

Reduces buffer bloat and traffic latency while also improving the user experience.

cBR-8 implements PIE algorithm as defined in RFC 8033 as the downstream AQM algorithm.

PIE Algorithm

Proportional Integral Controller Enhanced (PIE) controls latency by randomly dropping packets when latency is approaching or exceeding latency target. PIE is the AQM algorithm that is implemented on the cBR-8.

Explicit Congestion Notification (ECN) Marking

If packets are classified to Low Latency Service Flow (LL SF) and marked by the sender as ECN Capable Transport (ECT1), The CMTS may mark the Congestion Experienced bit (CE) if queue starts to build up Low Latency Low Loss Scalable Throughput (L4S) is a new technology that relies on ECN to provide high throughput and low latency for IP traffic. Examples of L4S congestion control algorithms are TCP Prague, Google BBR/BBR2, and Apple QUIC.

AQM Parameters

Table 2. SF AQM Disable

Type

Length

Value

25.40.1

1 

0   =
                Enable AQM on service flow 
1   =
                Disable AQM on service flow
Table 3. Downstream Classic AQM Latency Target

Type

Length

Value

25.40.1

1 

AQM Latency Target (in milliseconds) 

Recommended Latency Target Range: 10ms - 100ms

Configure Latency Target with service class: 

cable service class 300 downstream
cable service class 300 aqm-disable 0
cable service class 300 aqm-latency-target 15

Commands

Check AQM configuration

Router#show cable modem a84e.3f37.1740 verbose sup | i AQM
  AQM Disable                    : 0
  AQM Latency Target             : 12
  AQM Algorithm                  : 0
  AQM Max Threshold              : 2000
  AQM Range Exponent             : 19
  AQM Histogram Edge Count       : 15
  AQM Histogram Edges            : 0x00C8   2.00(ms)  0x0190   4.00(ms)  0x0258 6.00(ms)

Check AQM statistics 

Router#show cable dp aqm wideband-Cable 1/0/1:4
MAC Address      I/F       Sfid   class-name        Length/Max  Dequeues qDrops  rDrops   pDrops     Marked   Target
                                                                                                                                                                          Latency(ms)
a84e.3f37.18da   Ca1/0/1  17   6--class-map     0/255           0               0            0             0              0           12      
a84e.3f37.1740   Ca1/0/1  27   9--class-map     0/255          227968      0            0              0             4124     12  
a84e.3f37.102e   Ca1/0/1  33   12--class-map   0/255           0               0            0              0              0          12 

pDrops: policer drops
rDrops: random drops
qDrops: tail-drops + random drops
Marked: CE marked