100500UPLINK · Mbit/s · one dot per 60 sALERT ABOVE 90%POLL EVERY 60 sGetRequest · UDP 161ifHCOutOctets · port 21.3.6.1.2.1.31.1.1.1.10.2= 60 Mbit/s60% FULLALERT: > 90%PORT 7 DOWNTRAP: linkDownSEVERITY 3 · ERRORTOP: 192.168.1.20community "public"SIGNED + ENCRYPTEDPUSH EVERY 10 s
Coloured parts: the ones the current step is about; the rest wait in whiteAmber envelopes: SNMP requests · teal / indigo: replies from the router / switchCoins: an alert (bell), a trap, log lines, a flow record, streamed updatesBig screen: one dot per poll; the dashed line is the alert thresholdA monitoring server (192.168.1.5) polls the router and the switch with SNMP every 60 s: a GetRequest to UDP 161 asks for a counter named by an OID from a MIB, and the difference of two readings gives a rate (60 Mbit/s). Each rate becomes a dot on a graph; a threshold (90% for 3 polls) raises an alert to the on-call admin. Devices also speak first: an SNMP trap to UDP 162 when a port goes down, and syslog lines to UDP 514. Flow records (NetFlow v9, IPFIX) say who talked to whom: the backup PC used 84 of 96 Mbit/s. SNMPv3 signs and encrypts; streaming telemetry pushes counters every 10 s. The 60-second poll, the 90% for 3 polls rule, the 10-second stream, the counter readings and the graph are illustrative; the arithmetic on them is exact (450 000 000 bytes × 8 ÷ 60 s = 60 Mbit/s). The office uses 192.168.1.0/24 (RFC 1918); the cloud backup server 203.0.113.10 is an RFC 5737 documentation address. As the router sees LAN traffic before NAT, its flow records show the PC's private address.

Why watch

  1. A monitoring server keeps watch on the network. Every minute it asks the router and the switch for their numbers and draws them on a live graph. When something goes wrong, a person knows first, before users start complaining.

SNMP polling

Graphs and alerts

Devices speak up

Who, and what next

On the wire

;; step 1/11 · watch · poll
;; Monitoring server → Edge router
monitoring srv   192.168.1.5
watches          router 192.168.1.1, switch 192.168.1.2
every            60 s (illustrative)
collects         counters, traps, syslog, flow records
shows            graphs, alerts

Here one monitoring server is the SNMP manager, the flow collector and the syslog server. Bigger networks split these roles across several machines. Counters wrap back to zero when they overflow: a 32-bit byte counter wraps after 4 294 967 296 bytes, about every 344 s at 100 Mbit/s, which is why fast links are polled through the 64-bit "HC" counters (RFC 2863).