NetOp detects anomalies two different ways, on two different clocks. Understanding which one applies explains most of the "why didn't this alert?" and "why did this alert?" questions you will encounter.
Threshold anomalies: evaluated on the rolling hour
Every metric has one or more configured thresholds, each with its own severity — typically an early-warning band and a hard ceiling.
What is measured is the time the metric spent beyond the threshold within the last hour, not a single instantaneous spike. This is why the platform does not chase transients: a two-second blip accumulates almost nothing, while a sustained condition accumulates steadily.
It is also why observation text reads the way it does — for example, that a metric was above 70% for a cumulative duration of three minutes within the past hour. That sentence is the whole detection model. The approach is deterministic and explainable: you can always say exactly why an anomaly fired.
Real-time events: raised immediately
Device down, VPN down, and port down are detected in real time rather than waiting for the hourly evaluation. They raise an anomaly and open an incident immediately, and notify whoever is subscribed to the Operation category.
No duration accumulates first — the condition either exists or it does not, so there is nothing to average.
Client ports are the deliberate exception. An interface-down event on a port marked as user-facing is not reported at all, which is correct behaviour: users unplug things.
What NetOp watches
Resource — CPU (typically high at 70–80%, critical at 90%), memory (high at 60–70%, critical at 80%), and session counts, which should be checked against the vendor's supported maximum.
Connectivity — device down as a real-time anomaly, port errors counted per five-minute interval, and throughput as a percentage of configured bandwidth.
Experience — latency as round trip to the SLA target, jitter folded into SLA and VPN QoE, and packet loss as an average percentage to the SLA target.
Service — SLA QoE computed from latency, loss, and jitter; VPN metrics on the same model; and wifi errors covering association, authentication, DHCP, and DNS.
Always set two thresholds, not one
A single ceiling tells you only that damage is already happening. The lower band is what gives your team an hour of warning — the difference between an outage and a maintenance task. You can configure several lines on the same element, so an early-warning band and a hard ceiling coexist on one metric.
Understanding the SLA QoE score
NetOp combines latency, packet loss, and jitter into a single quality score between 1 and 5, calculated in real time per SLA profile. You do not have to reason about how three metrics interact — the score already does.
Scores fall into bands from poor through low, medium, and high to best. The exact band boundaries shift depending on the traffic type the profile is tagged as, so read them for the profile in front of you rather than memorising a set.
QoE profiles come from the SLAs defined on the firewall or the NetOp probe. NetOp monitors every configured profile automatically — you do not create them in the platform. The same model covers wired and wifi, which makes it the one score worth quoting in a service review.
Tag the profile, or the score is wrong
QoE is calculated against the expectations of an application type. A profile left untagged, or tagged as the wrong type, will score web browsing traffic by voice standards or the reverse — and report a network as failing when it is not.
This is the most common root cause of "the platform says our network is bad but it feels fine." Check the tagging under Settings › QoE before you investigate the network itself.
Related articles
- Anomalies, observations, and incidents
- Reducing alert noise
Comments
0 comments
Please sign in to leave a comment.