ENGINEERING TOPIC HUB
Network Monitoring & Observability
Choose network monitoring signals, compare SNMP, syslog, flow export, telemetry and probes, and build a source-backed response and retention workflow.
Start with a question, follow the evidence, then choose how to collect and respond. No collector, calculator or backend runs here.
Reviewed 2026-10-16 · 简体中文
What do you need to observe?
Need fault alerts?
Need interface health?
Need flow analytics?
Need telemetry?
Need long-term capacity trend?
Monitoring Data Retention Interface Utilization Related Tools
Reading rule: protocol facts cite primary sources. Question routes, review checklists and the worked design are editorial workflow suggestions. Polling intervals, alert thresholds, optical limits and retention duration require your platform/service policy; no universal numbers are invented.
Compare evidence types before choosing a method
This table summarizes the cited sections below; combinations are editorial suggestions, not an exclusive protocol ranking.
| Method | Purpose and output | Combine and validate |
|---|---|---|
| SNMP Polling | Use request/response retrieval for named management objects. GetBulk can retrieve groups of values; object definitions determine what each value means. This suits repeated state/counter observations. | Repeated retrieved values; interpret object definitions. |
| SNMP Trap | Use notifications to learn about reported transitions without waiting for the next scheduled read. SNMPv2 Trap has no protocol confirmation; an Inform requests a response. They are distinct notification behaviors. | Event notification; pair with current-state observations. |
| Syslog | Use structured device event messages to explain state changes and configuration activity. RFC5424 carries fields such as severity, timestamp and structured data; message format and transport are separate layers. | Correlate with other observations and verify collection health. |
| NetFlow | Use flow records to attribute observed traffic by exported keys and counters. NetFlow v9 uses templates to describe record layout; this section concerns v9, not every NetFlow version or vendor extension. | Traffic context; preserve exporter and sampling/model metadata. |
| sFlow | sFlow combines packet sampling with counter observations. A sampled packet view helps inspect traffic mix while counter samples describe monitored interfaces; keep these evidence types separate. | Traffic context; preserve exporter and sampling/model metadata. |
| IPFIX | IPFIX standardizes exchange of flow information between Exporting and Collecting Processes. Templates define record fields and Observation Domain scopes their interpretation. | Traffic context; preserve exporter and sampling/model metadata. |
| Streaming Telemetry | A subscription can deliver selected modeled paths repeatedly. OpenConfig gNMIc documents STREAM modes sample, on-change and target-defined; sample interval is a configured subscription attribute. | Correlate with other observations and verify collection health. |
| API Monitoring | Use an authenticated data API when relevant operational objects are exposed. RESTCONF is an HTTP-based interface for YANG-defined data; read operations and error responses have explicit semantics. | Correlate with other observations and verify collection health. |
| ICMP / Latency | ICMP Echo request/reply provides an IP-layer response signal. Track round-trip observations from a documented vantage point; this tests that exchange, not an application transaction. | Correlate with other observations and verify collection health. |
Worked architecture: branch users report slow backups
Illustrative design, not a diagnosis or a deployment configuration.
- Question: is the WAN busy, and which observed traffic contributes? Define one egress interface, service owner and common time window.
- Signals: collect comparable interface counters/state; add configured flow export for contributor context and syslog/notifications for transitions. Record exporter, direction, units, clocks and collection gaps.
- Correlation: align the slowdown window with utilization and top observed contributors. Check member distribution and errors before assuming aggregate load explains the symptom.
- Response: an operator-owned runbook requests corroborating service evidence, assigns the link/service owner and repeats the same service test after remediation. No fixed trigger is implied.
- Retention assumption: keep enough appropriate-resolution evidence for your incident review and capacity trend horizon under the approved privacy/access/deletion policy. Validate that queries survive aggregation.
- Follow-up: copy compatible manually observed rates into the Application Bandwidth Analyzer; then use explicit growth assumptions in the existing forecast planner. The hub does not collect or calculate these observations.
Monitoring Architecture Overview
Start with a service question and an inventory of devices, interfaces and responsible teams. Collection feeds normalization, time-series/event/flow storage, correlation and a response workflow. These are proposed design stages, not a required product stack.
Evidence and architecture role
Keep source identity, timestamps, observation direction and units alongside each signal. Pair state/counter observations with event context and flow attribution; one source alone is not a full service explanation.
Operational constraints and review
Operator checklist: define missing-data handling, collector ownership and credential boundaries before adding devices. Validate supported protocols on the actual platform; this hub deploys no collectors.
Source / policy: S-SNMP S-SYSLOG S-IPFIX S-EDITORIAL
SNMP Polling
Use request/response retrieval for named management objects. GetBulk can retrieve groups of values; object definitions determine what each value means. This suits repeated state/counter observations.
Evidence and architecture role
Record object identity, response time and observation time, not only the numeric value. Pair interface octet counters with discontinuity context before treating differences as traffic.
Operational constraints and review
Choose intervals and retry budgets from device capacity and operational needs; no universal interval is supplied. Use a supported authenticated/private SNMPv3 policy and verify implementation/security settings, rather than treating a version label as sufficient.
Source / policy: S-SNMP S-IF S-SECURITY S-EDITORIAL
SNMP Trap
Use notifications to learn about reported transitions without waiting for the next scheduled read. SNMPv2 Trap has no protocol confirmation; an Inform requests a response. They are distinct notification behaviors.
Evidence and architecture role
Retain sender identity, notification object and received time; then check current device/interface state. A notification describes a reported event, not continuous health.
Operational constraints and review
Operator response: test receiver reachability and planned failure events, deduplicate repeats and reconcile with polling. Silence alone must not be interpreted as proof that nothing changed.
Source / policy: S-SNMP S-EDITORIAL
Syslog
Use structured device event messages to explain state changes and configuration activity. RFC5424 carries fields such as severity, timestamp and structured data; message format and transport are separate layers.
Evidence and architecture role
Correlate origin and event time with interface state or flow changes. Preserve the original message for investigation while normalizing selected fields for search.
Operational constraints and review
Operator choices: validate clock quality, parser behavior, delivery transport and access to sensitive log fields. A severity number alone does not define service impact or the paging recipient.
Source / policy: S-SYSLOG S-EDITORIAL
NetFlow
Use flow records to attribute observed traffic by exported keys and counters. NetFlow v9 uses templates to describe record layout; this section concerns v9, not every NetFlow version or vendor extension.
Evidence and architecture role
Check exporter identity, observation interface/direction, template availability and sampling metadata before comparing contributors. Record the exporter configuration with the report.
Operational constraints and review
Operator review: missing templates or observation coverage can change what a collector can interpret. Exported flow records are not a complete packet capture or proof of application payload semantics.
Source / policy: S-NETFLOW S-EDITORIAL
sFlow
sFlow combines packet sampling with counter observations. A sampled packet view helps inspect traffic mix while counter samples describe monitored interfaces; keep these evidence types separate.
Evidence and architecture role
Save sampling parameters, agent and interface identity. Label derived traffic estimates as sampled estimates and compare their measurement boundary with interface observations.
Operational constraints and review
Operator caveat: a missing sample is not evidence that a short flow never existed. Validate hardware support and sampling configuration; do not present a sample set as an exhaustive packet history.
Source / policy: S-SFLOW S-EDITORIAL
IPFIX
IPFIX standardizes exchange of flow information between Exporting and Collecting Processes. Templates define record fields and Observation Domain scopes their interpretation.
Evidence and architecture role
Retain exporter/Observation Domain identity, templates and observation metadata together with records. Validate which Information Elements your exporter and collector actually support.
Operational constraints and review
Operator checks: inspect transport, sequence and template handling plus loss/collector health before calling a report complete. Standards-based export does not by itself ensure identical platform feature coverage.
Source / policy: S-IPFIX S-EDITORIAL
Streaming Telemetry
A subscription can deliver selected modeled paths repeatedly. OpenConfig gNMIc documents STREAM modes sample, on-change and target-defined; sample interval is a configured subscription attribute.
Evidence and architecture role
Preserve target, path, timestamp and subscription settings with observations. Define how an initial state, later updates and stale streams appear in dashboards.
Operational constraints and review
Operator validation: negotiate supported paths/encodings and test reconnects, collector backpressure and missing updates. Streaming is not a guarantee of unlimited frequency or universal model support.
Source / policy: S-GNMI S-EDITORIAL
API Monitoring
Use an authenticated data API when relevant operational objects are exposed. RESTCONF is an HTTP-based interface for YANG-defined data; read operations and error responses have explicit semantics.
Evidence and architecture role
Retain endpoint identity, schema/version, HTTP outcome and data freshness. Distinguish an empty supported dataset from authorization errors or an unavailable endpoint.
Operational constraints and review
Operator policy: use supported read permissions, verify HTTPS identities and handle pagination/rate limits as documented by that API. A successful HTTP response is not automatically evidence that a monitored service works.
Source / policy: S-API S-EDITORIAL
ICMP / Latency
ICMP Echo request/reply provides an IP-layer response signal. Track round-trip observations from a documented vantage point; this tests that exchange, not an application transaction.
Evidence and architecture role
Keep target, source location, packet settings and timestamps with loss/round-trip summaries. Compare observations before and after a reported event using the same probe definition.
Operational constraints and review
Operator caveat: investigate probe reachability and policy when responses disappear; corroborate with service checks. A round trip alone does not isolate forward versus reverse path delay.
Source / policy: S-ICMP S-EDITORIAL
Interface Utilization
Interface counters describe a specific interface/sub-layer. RFC2863 defines high-capacity octet counters and discontinuity context; choose comparable successive observations before deriving rates.
Evidence and architecture role
Arithmetic definition: rate bit/s = 8 × octet difference ÷ elapsed seconds; utilization % = rate ÷ documented capacity ×100. Handle resets/discontinuities before subtraction; preserve direction and actual elapsed time.
Operational constraints and review
Operator review: inspect errors/discards and member-level distribution alongside aggregate load. A long average can hide shorter changes; capacity and measurement layer must match. No universal safe utilization threshold is defined here.
Source / policy: S-IF S-EDITORIAL
CPU / Memory
Interpret resource values using their object definitions. HOST-RESOURCES-MIB defines processor load over a stated interval and storage entries with type, allocation units, size and used values.
Evidence and architecture role
Capture object/vendor model and time window with each series. Distinguish memory pools and storage types before constructing a used percentage; avoid comparing unlike denominators.
Operational constraints and review
Operator response: correlate resource changes with routing churn, configuration activity and collector gaps, then follow platform diagnostics. Do not infer a universal fault percentage or treat every device as implementing this MIB.
Source / policy: S-HOST S-EDITORIAL
Optical Power Monitoring
Read optical values with units, scale, status and freshness. Cisco ASR900 DOM documentation is a vendor/platform-specific example exposing transmit/receive power and other transceiver parameters with warning/alarm tables.
Evidence and architecture role
Record optic model, port/lane, supported sensor and both endpoints where available. Sensor identity and update context help distinguish unavailable readings from a low measured value.
Operational constraints and review
Operator review: use the actual optic/platform specifications and calibration context, then correlate with errors and link state. No fixed dBm limit is transferred across optics. Verify DOM support; the cited platform has documented operational limits.
Source / policy: S-SENSOR S-OPTICS S-EDITORIAL
Alert Design
Separate event arrival from an actionable alarm lifecycle. RFC8632 identifies alarms by resource/type and distinguishes alarm state/severity from operator management; delivery alone is not the response process.
Evidence and architecture role
Proposed runbook fields: resource, service impact evidence, owner, corroborating signal, acknowledgement, recovery check and escalation path. Keep maintenance suppression auditable and distinguish missing data from recovery.
Operational constraints and review
Operator policy: choose thresholds, persistence and hysteresis from your service objectives and observed baseline, then test detection and recovery. This hub provides no fixed paging threshold or automated severity score.
Source / policy: S-ALARM S-EDITORIAL
Monitoring Data Retention
Retention is an operator/data-governance decision, not a universal protocol duration. Decide which questions require raw events/flow records and which can use aggregated trends before selecting storage tiers.
Evidence and architecture role
Proposed worksheet: dataset, owner, resolution, access group, investigation/growth horizon, deletion rule and restore test. Keep exporter/schema context with retained records; avoid storing identifiers without a defined need.
Operational constraints and review
Operator trade-off: aggregation changes what later questions can be answered; document the transformation and validate representative queries. Choose duration from incident review, capacity planning, privacy obligations and actual storage budget; no fixed days are prescribed.
Source / policy: S-EDITORIAL S-IPFIX S-SYSLOG
Monitoring roadmap — planned only
These are scoped future ideas, not working tools. No page, tool number or active registration is created.
- SNMP Polling Load Calculator — Planned; not available
- Telemetry Data Rate Calculator — Planned; not available
- NetFlow Storage Estimator — Planned; not available
- Monitoring Server Sizing — Planned; not available
- Alert Threshold Planner — Planned; not available
Primary sources and claim boundaries
Accessed 2026-10-16. RFCs describe protocol/data semantics; vendor and software documentation are explicitly scoped to those implementations. Editorial advice is labeled separately.
- RFC 3416 — SNMP operations
S-SNMP · 4.2.1, 4.2.3, 4.2.6, 4.2.7 - RFC 2863 — Interfaces Group MIB
S-IF · 3.1.6; ifHCInOctets, ifHCOutOctets, ifCounterDiscontinuityTime - RFC 5424 — Syslog
S-SYSLOG · 4, 6, 8 - RFC 3954 — NetFlow v9
S-NETFLOW · 5–9 - RFC 3176 — sFlow
S-SFLOW · 2, 3 - RFC 7011 — IPFIX
S-IPFIX · 2, 3, 8, 10 - OpenConfig gNMIc — subscriptions
S-GNMI · Subscription configuration and stream modes - RFC 8040 — RESTCONF
S-API · 1, 2, 4.3, 7 - RFC 792 — ICMP
S-ICMP · Echo or Echo Reply Message - RFC 2790 — Host Resources MIB
S-HOST · hrProcessorLoad; hrStorageTable - RFC 3433 — Entity Sensor MIB
S-SENSOR · 3: sensor type, scale, precision, status and update time - Cisco ASR 900 — DOM for transceivers
S-OPTICS · Digital Optical Monitoring for Transceivers - RFC 8632 — Alarm management
S-ALARM · 3, 4, 5 - RFC 3414 — SNMPv3 USM
S-SECURITY · 1.4, 1.5
Related pages
EVPN control plane up, data plane down · EVPN CE ping failure case