When a single piece of infrastructure degrades, it rarely reports one problem. It reports dozens — across every vendor platform that touches it. This article explains how NetOp groups those related signals into a single Compound Incident, and how to read one when it lands in your queue.
Why one failure produces dozens of alerts
Threshold-based monitoring evaluates each device against its own limits, in isolation. It has no model of how your devices relate to one another, so it cannot tell the difference between five failures and one failure with four downstream effects.
Consider a core switch experiencing a transient fault. Within seconds, your monitoring stack may report all of the following:
- Access points reporting disconnected clients
- An SD-WAN controller reporting path degradation
- A cloud gateway logging application timeouts
- A firewall registering a throughput drop
Each of those is technically true. Each arrives in a different console. None of them names the switch. Your team opens multiple tickets and reconstructs the timeline by hand, and mean time to repair climbs while they do it.
What a Compound Incident is
A Compound Incident is a single record that groups co-occurring, causally related anomalies across vendors into one event. Rather than telling you that four things are broken, it describes the sequence: what failed first, what followed as a consequence, and which infrastructure was affected along the way.
Every Compound Incident separates two things that threshold alerting conflates:
- Root cause — the primary trigger the engine has isolated
- Symptoms — the cascading effects observed on other platforms
In the switch example above, the switch fault is the root cause and the remaining four alerts are attached as symptoms. You receive one item of work, not five.
How correlation works
NetOp sits as an aggregation layer above your existing hardware. It connects to your vendor platforms over their APIs and ingests event streams, logs, and telemetry in real time — it does not require agents on your devices or changes to your physical topology.
Correlation then runs in two stages:
- Topology mapping. NetOp builds a model of how your edge devices, core switches, and cloud instances connect. This model is what allows the engine to recognise that an access point anomaly and an SD-WAN anomaly share an architectural dependency.
- Relationship evaluation. Incoming anomalies are assessed against that model along with timing and behavioural context. Where the engine finds a dependency chain, it isolates the primary trigger and wraps the whole event into one actionable record.
Because the model is built from your environment rather than from generic rules, correlation accuracy improves as NetOp observes more of your network's normal behaviour.
What changes in your workflow
The practical difference is where your engineers spend investigation time. Instead of opening six dashboards to establish whether thirty alerts describe one problem or several, they open one incident that already answers that question and includes the event history and recommended remediation steps.
Two habits are worth unlearning during rollout:
- Do not close symptoms individually. Resolving the root cause resolves the Compound Incident. Closing attached symptoms by hand removes the feedback the engine uses to learn.
- Do not re-open per-vendor alerting in parallel. Running both models at once reintroduces the noise correlation exists to remove, and makes it hard to judge whether correlation is working.
When correlation gets it wrong
Correlation is probabilistic. Two genuinely independent failures that happen to coincide can occasionally be grouped, and a dependency NetOp has not yet observed can cause a related pair to arrive separately.
Both cases are worth reporting rather than working around. Resolution feedback is what reinforces the engine's understanding of your environment, so a corrected grouping improves subsequent accuracy. If a Compound Incident is grouped incorrectly, flag it on the incident record and note which signals should or should not have been included.
Related articles
- Operating a multi-vendor network with NetOp
- From dashboards to autonomous operations
Comments
0 comments
Please sign in to leave a comment.