In-Network INT Sampling and Aggregation for Link Latency Monitoring

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current methods for real-time network-wide link latency monitoring using in-network in-band network telemetry (INT) are costly and inefficient, as they require extensive data collection and processing, leading to network overload and potential biases in measurement results.

Innovation Solution

The proposed solution involves a low-cost approach to achieve real-time network-wide link latency monitoring by implementing in-network INT sampling and aggregation. At each egress port of a network device, packets are selected and INT headers and data are inserted only when necessary, ensuring network-wide coverage while reducing network processing load and data export packets.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If dedicated source-routing probing packets are injected with INT to traverse pre-determined paths, then network-wide coverage is achieved, but topology changes or link/node failures require recalculation and the injected packets consume extra bandwidth

Engineering Contradiction:
Improvenetwork-wide coverageVSAvoidextra bandwidth consumption
Core Design Contradiction:
ReliabilityVSQuantity of substance

Solution Approach 1:

User traffic packets themselves carry INT headers and perform telemetry measurements, eliminating the need for separate dedicated probing packets. The normal data traffic serves dual purposes: delivering user data and collecting network telemetry information, thus achieving network-wide coverage without consuming additional bandwidth for probing.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The INT header is made universal by enabling it on a sample of user traffic packets that traverse all links in the network. These multi-functional packets simultaneously serve as user data carriers and telemetry measurement vehicles, achieving both data transmission and network monitoring goals with a single packet type.

Inventive Principle:
Principle #6Universality (Multi-functionality)

2Reliability

If INT is enabled on all user traffic at network ingress, then network-wide coverage is achieved, but the cost is prohibitive due to extensive data collection and processing

Engineering Contradiction:
Improvenetwork-wide coverageVSAvoiddata collection and processing load
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

Instead of enabling INT on all user traffic packets (excessive action), the system enables INT on a carefully selected sample of packets that guarantees network-wide coverage (partial action). This sampling approach achieves the necessary monitoring coverage while significantly reducing the volume of telemetry data that needs to be collected, processed, and exported, thus lowering system complexity and processing costs.

Inventive Principle:
Principle #16Partial or excessive action

Solution Approach 2:

The system extracts only the essential telemetry information needed for network-wide coverage from the sampled packets, rather than collecting comprehensive data from all packets. By taking out only the necessary measurement data from a representative sample, the system achieves network-wide visibility with minimal data processing requirements.

Inventive Principle:
Principle #2Taking out (Extraction)

3Measurement precision

If 1-hop INT is enabled on each link with sampling at egress ports, then link latency data is collected, but too many data export packets are generated overwhelming the data collector

Engineering Contradiction:
Improvelink latency dataVSAvoidnumber of data export packets
Core Design Contradiction:
Measurement precisionVSQuantity of substance

Solution Approach 1:

The system merges telemetry data from multiple links into a single aggregated INT header that travels with sampled packets. Instead of generating separate export packets for each link's telemetry data, the Int headers are combined and carried together by the sampled user traffic packets, significantly reducing the total number of export packets to the data collector while maintaining measurement precision.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

Sampled user traffic packets serve as intermediaries that carry aggregated telemetry data from multiple network links to the data collector. These intermediary packets transport consolidated measurement information without requiring separate dedicated export packets for each link, thus reducing the overall packet volume to the collector while preserving link latency measurement accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentUS12323313B2Real-time network-wide link latency monitoring with in-network INT sampling and aggregation
Publication Date: 2025.06.03 HUAWEI TECH CO LTD
  • US12323313B2 patent drawing
  • US12323313B2 patent drawing
  • US12323313B2 patent drawing

AI summary

A method performed by a network device along a data path for real-time network-wide link latency monitoring. The method includes receiving a packet; incrementing a packet counter; determining whether the packet counter is at least equal to a silent period value (M) when a sampling cycle flag is not set; determining whether the packet includes an existing in-band network telemetry (INT) header when the packet counter is at least equal to the silent period value (M); inserting INT data into the existing INT header of the packet and setting the sampling cycle flag when the packet includes the existing INT header; and forwarding the packet along a data path towards a destination device.