Skip to content
Blog

What flow data tells you that interface counters cannot

SNMP graphs show that a link is busy. Flow telemetry explains why. A practical look at NetFlow and IPFIX for network operators.

E-Net Engineering

Every network team has an interface graph that turns red at 9 p.m. The graph is accurate: the link is saturated. What it cannot tell you is who is using it, where the traffic is going, or whether this is normal.

That is the difference between counters and flows.

Counters measure volume

SNMP interface counters report bytes and packets per interface. They are cheap to collect, universally supported and excellent for answering one question: how full is this link?

They are poor at every follow-up question. A 10 Gbps uplink at 95% utilization could be a video streaming peak, a single misconfigured backup job, or a volumetric attack. The counter looks the same in all three cases.

Flows describe conversations

A flow record summarizes a conversation between two endpoints: source and destination addresses, ports, protocol, interfaces, byte and packet counts, and timestamps. Routers and switches export these records using standard protocols:

  • NetFlow v5 / v9 — widely deployed, with v9 introducing templates for flexible fields.
  • IPFIX — the IETF standard evolved from NetFlow v9.
  • sFlow — packet sampling, common on switches.

With flow data, the 9 p.m. question becomes answerable: which prefixes are the top talkers, which applications dominate, and which upstream autonomous systems the traffic comes from.

Enrichment makes flows useful

Raw flow records are just addresses and numbers. They become operationally useful when enriched with context:

  • ASN — attributing traffic to networks informs peering and transit decisions.
  • Geolocation — shows where traffic originates and terminates.
  • Interface and site names — maps flows to physical topology.
  • Subscriber or tenant identity — connects traffic to the customers who generate it.

This enrichment is where most of the engineering effort in NetPulse goes. Collection is a solved problem; turning collected data into answers is not.

Sampling is fine for most questions

Operators sometimes hesitate because high-speed links require sampled flow export. For capacity planning, top-talker analysis and anomaly visibility, sampled data is statistically reliable. Unsampled export is worth the cost mainly where you need per-conversation accounting.

Keep counters, add flows

Flow telemetry does not replace SNMP. Counters remain the most efficient way to watch utilization. Flows answer the questions that come after the alert — and the historical questions that come weeks later during a capacity review.

Get started

Ready to see it on your own setup?

Tell us what you run today. An engineer will walk you through the right E-Net Communication product for your network, infrastructure or AI workload.