Modern Telemetry – Streaming vs Traditional
Learning Objective: Understand modern telemetry – the evolution from traditional SNMP polling to streaming telemetry (gNMI, gRPC, Kafka) for real-time, high-frequency network monitoring.
The Evolution of Network Telemetry
From "asking" devices for data to devices "telling" OSS when data changes or on fixed intervals.
Traditional Approach: SNMP Polling
SNMP (Simple Network Management Protocol) has been the dominant telemetry protocol for decades. OSS/NMS polls devices at fixed intervals (e.g., every 5 minutes) to collect metrics.
How SNMP Polling Works
- OSS requests data from devices (pull model)
- Devices respond with current values
- Polling intervals: 5 min, 15 min, hourly
- Data is sampled, not continuous
- Structured as MIBs (Management Information Bases)
Limitations of SNMP Polling
- Sampling gaps – Misses spikes between polls
- Limited scalability for modern networks – Originally designed for earlier network generations
- No real-time visibility – Latency = poll interval
- Overhead – Polling requests consume device CPU
- SNMP traps still coexist – Used for alarms, but traps alone don't provide continuous metrics
Traditional SNMP Polling Pattern
OSS: "What is ifInOctets on port 1?" Device: "1972837462 bytes" -- wait 5 minutes -- OSS: "What is ifInOctets on port 1?" Device: "1985000000 bytes"
Even in modern OSS environments, SNMP traps remain widely used for alarm and fault notifications. Most telecom operators run hybrid architectures combining SNMP polling, SNMP traps, and streaming telemetry together.
Modern Approach: Streaming Telemetry
Streaming telemetry uses a push model where devices continuously send data to OSS systems without waiting for requests. Enables real-time, high-frequency monitoring.
How Streaming Telemetry Works
- Devices push data to OSS (push model)
- Data sent on change or at sub-second intervals
- Real-time or near real-time (milliseconds)
- Streaming protocols: gNMI, gRPC, NETCONF
- Data encoded in protobuf, JSON, or GPB
Key Advantages
- Real-time visibility – Sub-second latency
- No sampling gaps – Every data point captured
- Scalable – Devices push once, many consumers
- More efficient – Reduces repetitive polling and enables event-driven updates
- Structured data – YANG models, protocol buffers
Two Common Streaming Telemetry Models
- Sampled telemetry: Device sends updates at fixed intervals (e.g., every 100ms, 1 second, 5 seconds)
- On-change telemetry: Device sends updates only when values change significantly (threshold-based delta reporting)
On-change telemetry reduces data volume while still capturing meaningful changes. Many deployments use both models for different metric types.
Streaming Telemetry Pattern (gNMI/gRPC)
Device (Streaming): "ifInOctets updated: 1972837462 bytes" -- milliseconds later -- Device (Streaming): "ifInOctets updated: 1973000000 bytes" -- continuously pushes every 100ms or on change --
Modern OSS platforms do not use telemetry only for dashboards. Real-time telemetry drives orchestration engines, closed-loop automation, anomaly detection, AI-based assurance, and intent-based networking decisions.
Key Telemetry Protocols Compared
| Protocol | Model | Data Format | Use Case |
|---|---|---|---|
| SNMP | Poll (request-response) | ASN.1/MIB | Legacy monitoring, alarms, traps |
| NETCONF | RPC-based configuration | XML/YANG | Primarily configuration management |
| gNMI | Streaming (push) + RPC | Protobuf/JSON + YANG | Modern streaming telemetry, config |
| gRPC | Streaming + RPC | Protobuf | Transport framework used by gNMI and microservices |
| RESTCONF | REST-based | JSON/YANG | HTTP-based configuration and telemetry |
gNMI – The Modern Standard for Streaming Telemetry
gNMI (gRPC Network Management Interface) is becoming the industry standard for streaming telemetry in cloud-native and 5G networks. It is defined by the OpenConfig working group.
gNMI Capabilities
- CAPABILITIES: Discover device capabilities
- GET: Retrieve current state
- SET: Update configuration
- SUBSCRIBE: Streaming telemetry (push)
Sample gNMI SUBSCRIBE
Subscribe Request: path: /interfaces/interface/state/counters mode: STREAM sample_interval: 100ms
Real-World Example: SNMP vs Streaming
A 5G gNB experiences a 2-second latency spike every 3 minutes:
- SNMP (polling every 5 minutes): Misses the spike entirely. Reports average latency = normal. Operator unaware.
- Streaming telemetry (100ms intervals): Captures every spike. NOC sees pattern. Identifies root cause → periodic interference.
Streaming telemetry reveals problems that SNMP sampling misses. Critical for 5G SLAs and real-time assurance.
YANG (Yet Another Next Generation) is a data modelling language used to define telemetry data structures. OpenConfig and IETF YANG models standardize telemetry across vendors. gNMI uses YANG models to represent device state and configuration.
Modern Telemetry Pipeline
Streaming data flows through message buses for real-time processing, analytics, and automation.
Why Telemetry Matters in Real Operations
- 5G services and network slices increasingly require near real-time monitoring and assurance.
- Proactive operations: Detect anomalies before customers report them
- Cloud-native networks: Dynamic, ephemeral resources need streaming telemetry
- Cost optimization: Precise capacity management with real-time data
- Automation: Streaming data triggers closed-loop orchestration
- Multi-vendor consistency: OpenConfig YANG models normalize across vendors
- Legacy devices: Many devices don't support gNMI or streaming telemetry
- Hybrid architectures: Operators run both SNMP and streaming in parallel
- Data volume: Streaming generates massive data requiring Kafka and time-series databases
- Skills gap: Teams need training on gNMI, YANG, protobuf, Kafka
- Mediation layers: Legacy EMS must adapt to streaming telemetry
Real-time streaming telemetry feeds network digital twins – virtual replicas of physical networks used for simulation, what-if analysis, and predictive maintenance.
Connection to BSS
- SLA dashboards: Real-time performance data for customer-facing SLAs
- Usage-based billing: High-frequency usage data for dynamic pricing
- Customer experience: Real-time QoE data for premium services
- Predictive maintenance: Streaming telemetry enables proactive repairs before customer impact
Common Interview Questions
Q1. What is the difference between SNMP polling and streaming telemetry?
SNMP uses a pull model – OSS requests data at fixed intervals. Streaming telemetry uses a push model – devices continuously send data without waiting for requests. Streaming enables real-time visibility without sampling gaps.
Q2. What are the two common streaming telemetry models?
Sampled telemetry (fixed interval updates) and on-change telemetry (updates only when values change significantly).
Q3. What is gNMI and why is it important?
gNMI (gRPC Network Management Interface) is a modern protocol for streaming telemetry and configuration management. It is becoming the industry standard for cloud-native and 5G networks.
Q4. What are YANG models?
YANG is a data modelling language that defines the structure of telemetry data. OpenConfig YANG models standardize telemetry across vendors, enabling multi-vendor OSS integration.
Q5. Why is streaming telemetry critical for modern OSS?
It powers real-time assurance, closed-loop automation, anomaly detection, AIOps, and orchestration – not just dashboards.
Q6. What are the challenges of migrating from SNMP to streaming telemetry?
Legacy device support, hybrid architecture complexity, massive data volume, skills gap, and adapting existing EMS/NMS systems.
Key Terms
Takeaways for You
- SNMP polling = OSS requests data (pull). Sampling gaps, limited scalability, still widely used.
- Streaming telemetry = Devices push data (push). Real-time, no gaps, more efficient.
- Two streaming models: Sampled (fixed interval) and on-change (event-driven).
- gNMI is the modern standard for streaming telemetry (gRPC + YANG).
- YANG models define telemetry data structures (OpenConfig for multi-vendor standardization).
- 5G services increasingly require near real-time monitoring and assurance – streaming telemetry is essential.
- Telemetry powers automation – not just dashboards, but orchestration, AIOps, and closed-loop control.
- SNMP traps still matter – most operators run hybrid SNMP + streaming architectures.
- Streaming data flows through Kafka/Pulsar to analytics engines and AIOps platforms.
- Digital twins are emerging from real-time streaming telemetry for simulation and prediction.
Recommended Next Learning Path