Overview of Telemetry

Definition

Telemetry is a technology that remotely and quickly collects data from physical or virtual devices. Devices periodically send information (such as interface traffic statistics, CPU usage, and memory usage) to collectors in push mode. The push mode collects data faster than the conventional pull mode (question-answer interaction).

Purpose

As the number of devices on the SDN network continuously increases with an increasing number of services carried, users pose higher requirements on intelligent O&M for the SDN network. Specifically, data monitoring requires sampling with a higher precision, so that microburst traffic can be efficiently detected and adjusted. In addition, the impact of monitoring on device functions and performance must be minimal to improve device and network utilization.

Conventional network monitoring methods (such as SNMP get and CLI) have low management efficiency and cannot meet user requirements due to the following disadvantages:

The pull mode is used to obtain devices' monitoring data. This mode however cannot monitor a large number of network nodes, limiting network growth.
The interval for sampling monitoring data is in minutes. To improve accuracy, data must be queried more frequently, which increases the CPU usage of network nodes and affects normal device functions.
Data obtained from monitored network nodes is inaccurate due to network transmission delays.

A new network monitoring mode is required to implement large-scale and high-performance monitoring on network devices. Telemetry has therefore been developed to meet this requirement. With this technology, the intelligent O&M system can manage more devices, monitoring data can be obtained in real time with higher precision, and the monitoring process exerts minimal impact on device functions and performance. It also provides the most important big data basis to quickly locate faults and optimize the network quality. It converts network quality analytics into big data analytics, effectively supporting intelligent O&M.

Table 1 compares telemetry and conventional network monitoring methods.

**Table 1** Comparison between telemetry and conventional network monitoring methods
Item	Telemetry	SNMP get	SNMP trap	CLI	Syslog
Working Mode	Push	Pull	Push	Pull	Push
Precision	Subseconds	Minutes	Seconds	Minutes	Seconds
Structured or Not	Structured using the YANG model	Structured using MIBs	Structured using MIBs	Unstructured	Unstructured

As described in Table 1, although SNMP trap and syslog use the push mode, only traps or events are pushed. Monitoring data such as interface traffic statistics cannot be collected or sent.

Telemetry provides two subscription modes based on the roles of devices and collectors: static subscription and dynamic subscription. Table 2 compares the two modes.

**Table 2** Comparison between telemetry static and dynamic subscription
Item	Static Subscription	Dynamic Subscription
Device and collector roles	The device functions as the client, and the collector functions as the server.	The device functions as the server, and the collector functions as the client.
Subscription process	The device initiates a connection to the collector to collect and send data.	The collector initiates a connection to the device and collects data by listening to the device.
Subscription mode	Based on the UDP or gRPC dial-out mode	Based on the gRPC dial-in mode
Application scenario	Long-term, periodic monitoring of device status	Short-term monitoring of device status