Streaming network telemetry is a real-time data collection service in which network devices such as routers, switches and firewalls continuously push data related to the network's health to a centralized location. In general, telemetry is the science of transmitting measurements from a remote source to a receiving station for storage and analysis.
Data streaming network telemetry is push-based and data is transmitted automatically and continuously. The collected data from streaming can then be used to:
- Set up network monitors and alerts based on pre-configured thresholds.
- Create network performance baselines.
- Plan network capacity requirements.
- Troubleshoot connectivity and performance issues.
- Use artificial intelligence (AI) to make automated decisions.
How streaming network telemetry works
Data is streamed with protocols such as TCP, UDP, XML-RPC or gRPC in one of two ways. The first method is to send information to a collector at regularly-configured intervals. This is known as cadence-based telemetry. This helps build historical baselines. The second method is to send data when a specific event occurs – such as a critical link going down or the throughput on a link passes a specified maximum threshold. This is referred to as event-based telemetry.
Streaming network telemetry vs. SNMP
Prior to streaming network telemetry technologies, the simple network monitoring protocol (SNMP) was the standard method for collecting network telemetry information. However, SNMP uses a pull technology where a SNMP collector had to request telemetry information at regular intervals. Within typical enterprise networks, a collector would be configured to request information between 5 and 30 minutes. Thus, administrators end up with time gaps in the information gathered. Additionally, polling is fairly resource intensive and does not scale well.
Streaming network telemetry, on the other hand, is a push technology. Instead of having the collector request data every few minutes, the network device will stream health and interface data in near-real time. Ultimately, it can provide more data point granularity with better performance when compared to SNMP.
Architecture and framework of streaming network telemetry
Network components that are capable of streaming network telemetry information can be configured to point to one or more data collectors. The communications sessions can be established by the network device subscribing to the collector, or the collector subscribing to the network device.
Streaming network telemetry can alert networking professionals about unusually sharp increases or decreases in traffic. The data collected by streaming telemetry is model-driven and ideal for using big data analytics to optimize traffic flow.
From an implementation standpoint, administrators have control over several aspects of the subscription and stream process. First, they choose who initiates the subscription conversation, either the network device or the collector. Next, an administrator can selectively choose what information they want to collect. This process is accomplished using standard data models such as IETF YANG, OpenConfig or vendor proprietary models. Finally, admins choose the frequency that cadence-based data is streamed to a collector.