Home › Topics › Network Monitoring Fundamentals › Glossary

Network Monitoring Terms Defined: the A-to-Z Reference for People Who Have a Network to Run


Every discipline has a vocabulary that newcomers are expected to absorb by osmosis, and network monitoring's is worse than most because half of it was coined by vendors. This glossary defines the terms you will meet in a monitoring console, a vendor data sheet or a 3 a.m. alert, in a few sentences each, with a link to the guide that goes deeper. It is alphabetical within sections: the metrics first, then the protocols, then the practice of monitoring itself. If you want fewer terms with worked examples instead of a reference, the working glossary of 24 monitoring terms is the companion piece.


Jump to: Metrics and what they mean Protocols and data sources The practice of monitoring Devices, services and infrastructure


Metrics, and what the numbers mean


Bandwidth (in a computer network)

The maximum rate at which a link can carry data, in bits per second: a gigabit Ethernet port has 1,000 Mbps of bandwidth whether or not anyone is using it. In everyday use "bandwidth" also means how much of that capacity is in use, which is really utilization. The core network metrics series keeps the two apart.

Bandwidth, internet (what "bandwidth" means on your bill)

The rate your provider sells, usually quoted downstream and upstream separately, and the ceiling on everything that crosses the WAN. The number to compare it with is your measured peak utilization, not your average.

Bandwidth utilization (and the formula)

Bytes transferred in an interval, multiplied by eight, divided by the interval in seconds, divided by the link speed, as a percentage. Interface counters give the bytes; the interval is your polling period. Above about 80% sustained, queues build and latency rises.

Bandwidth hogging (and the bandwidth hogger)

One host, application or conversation consuming a disproportionate share of a link. The interface counter shows that the link is full; flow records (NetFlow, sFlow, IPFIX) show who the hogger is. The top talkers series covers finding them.

Bandwidth management and bandwidth allocation

Deciding, usually in a router or firewall, which traffic gets how much of a link: rate limits, QoS classes, priority queues. Monitoring tells you whether the allocation matches what the business actually needs.

Bandwidth conversion (bits, bytes, Mbps and MB/s)

Links are rated in bits per second; files and counters are measured in bytes. Divide Mbps by eight for MB/s: a 100 Mbps link moves 12.5 MB per second at best. Most "the link is slower than advertised" complaints are this arithmetic.

Bandwidth increase (when to upgrade)

The decision point is sustained peak utilization, not the monthly average. A link that averages 20% but sits at 95% every morning from nine to ten needs more bandwidth, or a schedule change. The capacity planning series turns history into the decision.

Throughput

The rate actually achieved by a transfer, after protocol overhead, loss and latency have taken their share. Throughput is always less than bandwidth; how much less is the diagnostic. The latency versus bandwidth guide explains why a fast link can still feel slow.

Latency (in a computer network)

The time a packet takes to get from one point to another, in milliseconds, usually measured as a round trip. Latency comes from distance (the speed of light in fibre is about 200 km per millisecond), from queues in busy devices, and from each hop's processing. A latency issue is rarely about the distance and almost always about a queue.

Latency, higher (what higher latency means)

Longer waits for every request and reply, which matters most to interactive traffic: a voice call, a remote desktop, a database round trip. Higher latency with no packet loss usually means congestion somewhere on the path; higher latency with loss means a device is dropping what it cannot queue. Lower is better, and consistent is better than merely low.

Latency, loaded versus unloaded

Unloaded latency is measured on an idle link; loaded latency is measured while the link is busy. The difference is "bufferbloat": queues filling up. A link with 2 ms unloaded and 200 ms loaded is the usual cause of "the internet is fine until someone uploads something."

Jitter (in computer networks)

The variation in latency from packet to packet. A steady 40 ms is fine for a voice call; 40 ms that swings between 10 and 90 is not, because the receiver cannot play audio that arrives out of rhythm. Jitter is the metric for real-time traffic, and the jitter and real-time traffic series covers measuring and fixing it.

Packet loss (meaning, and what counts as bad)

The share of packets sent that never arrive, as a percentage. TCP hides small amounts of loss by retransmitting, at the cost of throughput; real-time protocols cannot. Packet loss of 1% is noticeable on a voice call and bad for video; anything sustained above 0.1% on a wired network is a fault to find. The packet loss series covers where packets die.

Packet loss detection

Measuring loss, either actively by sending test packets (ICMP pings, or a packet train between two probes) or passively from interface error and discard counters. A packet loss detector that runs continuously catches the intermittent loss that a one-off test misses.

Round-trip time (RTT)

The latency there and back, which is what ping reports and what most monitoring measures, because one-way latency needs synchronised clocks at both ends.

Uptime (definition)

The time a device or service has been running or reachable, either as a duration since the last restart (what the device itself reports) or as a percentage of a period during which it was available. The two meanings get confused constantly; a server with 400 days of uptime and a service that was down for an hour last Tuesday are both "uptime" numbers.

Availability (and how it is calculated)

The percentage of a period during which something was reachable and working: available time divided by total time, where total time should exclude the moments the monitor was not looking. 99.9% ("three nines") allows about 8.8 hours of downtime a year; 99.99% allows about 53 minutes. The uptime and availability series explains what "up" has to mean for the number to be trusted.

Downtime

Time during which something was unavailable, whether planned (a maintenance window) or not. Honest availability reporting counts the unplanned kind and labels the planned kind separately.

Error rate (interface errors and discards)

The share of frames on a port that arrived damaged (errors) or were thrown away because the queue was full (discards). A rising error rate on one port is a cable or a duplex mismatch; rising discards are congestion.

Response time

How long a service takes to answer a request: a web page, a DNS query, a database call. Measured by a monitor acting as a client, which is the only way to see what users see.

MTTD and MTTR

Mean time to detect and mean time to repair: how long a fault exists before anyone knows, and how long it takes to fix once known. Monitoring exists to shrink the first; the MTTD and MTTR series covers measuring both.


Protocols and data sources


SNMP (Simple Network Management Protocol)

The question-and-answer protocol monitoring systems use to read counters from network devices: a manager asks on UDP port 161, an agent on the device answers, and the device can send traps to UDP port 162. The what is SNMP guide is the full explanation.

MIB (Management Information Base) definition

The catalogue of everything an SNMP device can be asked about, organised as a tree and described in MIB files published by standards bodies and vendors. To define MIB in one line: the dictionary that turns a numeric OID into a name, a type and a unit.

OID (object identifier)

The dotted-number address of one object in the MIB tree, such as 1.3.6.1.2.1.2.2.1.10 for an interface's inbound byte counter. An OID lookup translates the number into its name using the MIB.

SNMP trap and inform

A message a device sends on its own when something happens (a trap), or the same with an acknowledgement so the device knows it arrived (an inform). Received by a trap receiver on UDP 162.

Community string

The shared password of SNMP v1 and v2c, sent in clear text. Change it from the default, make it read-only, and restrict which addresses may use it; or use SNMP v3.

WMI (Windows Management Instrumentation)

The built-in instrumentation layer of Windows, which a monitor queries remotely to read services, disks, CPU, memory, event logs and application counters without installing an agent. The WMI monitoring guide covers the ports and permissions it needs.

SSH-based monitoring

Reading a Linux or Unix host by logging in over SSH and running a command, whose output becomes the reading. The agentless equivalent of WMI for non-Windows hosts.

IPMI and Redfish

Protocols for talking to a server's management controller (BMC, iDRAC, iLO) through its dedicated management port: temperatures, fans, power and hardware health, available even when the operating system is down. The out-of-band management monitoring series covers them.

NetFlow

A protocol by which a router or switch exports a summary of every conversation it forwards (source, destination, ports, protocol, bytes, packets) to a collector. NetFlow v5 is the simple fixed format; v9 is template-based. The flow monitoring series starts from the beginning.

NetFlow analyser (analyzer) and flow collector

A flow collector receives the exported records; a NetFlow analyser turns them into top talkers, conversations, per-interface traffic and reports. The two words are used interchangeably for the same product, with the British and American spellings used interchangeably for the same word.

sFlow

A sampling-based alternative to NetFlow, common on switches: one packet in N is sampled and reported along with interface counters, which scales to very fast links at the cost of exactness. Sampling rates must be honoured by the collector or the totals are wrong.

IPFIX

The IETF standard descended from NetFlow v9: the same template-based idea, vendor-neutral, with extensible fields. "IPFIX versus NetFlow" is mostly a question of which your devices export.

Flow exporter

The device, or the interface on it, configured to send flow records to a collector. An exporter that stops sending is a silent gap in the traffic picture unless the collector raises an alert for the silence.

Top talkers

The hosts, conversations or applications sending the most traffic in a period, derived from flow records. The report that answers "what is eating the bandwidth."

Syslog

The protocol for sending one-line log messages from devices and hosts to a collector, by default over UDP port 514, with a severity from 0 (emergency) to 7 (debug) and a facility that says what sent it. The syslog ports and severity reference has the tables.

ICMP (and ping)

The Internet Control Message Protocol; an ICMP echo request and its reply are what "ping" sends and receives, giving reachability and round-trip latency in one cheap packet. The ping, traceroute and ICMP tools series covers what ping can and cannot tell you.

Traceroute

A tool that lists every hop between you and a destination with the latency to each, by sending packets with increasing time-to-live values. The path analysis series explains reading it honestly.

LLDP and CDP

Link Layer Discovery Protocol and Cisco Discovery Protocol: switches announce themselves to their neighbours on every port, and anything listening can draw the physical topology from the announcements. The source of an auto-generated network map.

ARP and the MAC address table

The tables that map IP addresses to hardware addresses and hardware addresses to switch ports; read over SNMP, they tell you which port a device is plugged into. The basis of device discovery and rogue-device detection.

Packet capture (pcap)

Recording the actual packets on a link for later analysis, as opposed to flow records, which summarise conversations. Heavy, exact and short-term; the packet capture analysis series covers when it is the right tool.

Port mirroring (SPAN)

A switch feature that copies the traffic of one or more ports to another port, where a capture or analysis tool can see it. How you get packets to a capture box without being in the path.

NTP

Network Time Protocol, which keeps every device's clock in step. Unsynchronised clocks make every log and every graph across devices impossible to correlate; monitoring NTP offset is cheap insurance.

DNS, DHCP and IPAM

The name service, the address-assignment service and the practice (or product) of tracking which addresses are allocated where: IP address management. An IPAM server records subnets, ranges and assignments; the IPAM and subnet monitoring series covers watching the three together.

Streaming telemetry

The newer push model in which a device streams its counters to a collector continuously instead of being polled; higher resolution, more data, fewer devices that support it. The streaming telemetry series covers where it fits beside SNMP.

Prometheus and OpenMetrics

A text format in which applications expose their own metrics on an HTTP endpoint for a collector to scrape. A monitoring tool that scrapes those endpoints reuses what developers already export instead of duplicating it.

API monitoring and webhooks

Watching a service through its programming interface, and sending alerts to other systems through theirs. The monitoring APIs and webhooks series covers both directions.


The practice of monitoring


Monitoring (versus observability)

Collecting known measurements from systems on a schedule and comparing them with expectations. Observability, by one common definition, is the property of a system whose internal state can be inferred from what it emits; in practice the word is used for the richer end of monitoring (metrics, logs and traces together) and for a market segment. The monitoring, observability and telemetry series translates the vocabulary.

Observability (definition)

A measure of how well you can understand what a system is doing from the outside, from its metrics, logs and traces. For most networks the honest version is "monitoring with enough context to answer why, not just what."

Telemetry

Measurements transmitted from a system to wherever they are stored and analysed. Polled counters, pushed metrics, flow records and log lines are all telemetry.

NMS (network management system)

A system that discovers devices, polls them, stores the history, draws the graphs and raises the alerts; the product category this library is about. An NMS reached through a browser is sometimes called a web NMS. The network monitoring fundamentals series explains what one is for.

NOC (network operations centre)

The team, or the room, that watches the monitoring and responds to it. A NOC dashboard or wall display is the always-on screen designed to be read from across the room.

Agent and agentless

An agent is software installed on each monitored host to report its state; agentless monitoring reads the host remotely through what it already exposes (WMI, SSH, SNMP, IPMI). Agentless means nothing to deploy or patch on the hosts.

Poller and polling interval

The component that asks each device for its readings, and how often it asks. Shorter intervals give finer graphs and cost more; the polling and granularity series covers choosing.

Sensor (and the sensor count)

One measured thing on one device: an interface's inbound traffic, a volume's free space, a service's state. Some products license by the number of sensors, which turns thoroughness into a cost; others license by device.

Threshold

The value at which a reading becomes a warning or a critical. Static thresholds are typed in; learned thresholds come from the sensor's own recent readings. The threshold tuning series covers setting them without crying wolf.

Baseline

What normal looks like for a metric, learned from history, so that "unusual" can be detected without a hand-set threshold. A network baseline should be established before the first big change and refreshed after it. The baselining series covers how.

Alert, notification and incident

An alert is the monitor's judgement that something crossed a line; a notification is the message that carries it to a person or a system; an incident is the record of a real problem, which may collect many alerts. The alert design series covers keeping the three apart.

Alert fatigue

What happens when a monitoring system sends so many notifications that people stop reading them, and then miss the one that mattered. The alert fatigue series is the cure.

Alert storm

Dozens or hundreds of alerts from one cause, typically every device behind a failed switch or link. Dependency-aware suppression turns the storm into one incident.

Dependency (parent and child)

The relationship that says device B is reachable only through device A, so that when A fails, B is marked unreachable rather than down. The topology, used for alerting.

Down versus degraded

Down means unreachable; degraded means reachable but with a problem (a full disk, a stopped service). Keeping them separate keeps the availability record honest and the pager quiet.

Flapping

A state that changes repeatedly in a short time (up, down, up, down), producing an alert each way. Debounce and repeat-count damping stop a flapping device paging anyone.

Debounce (duration and repeat-count damping)

Requiring a condition to persist for a duration, or across several polls, before it becomes an alert, so that a momentary spike pages nobody.

Escalation and on-call

The rule that an unacknowledged alert goes to the next person or channel after a set time, and the rota that says who that is. The escalation and on-call series covers designing both.

Maintenance window

A scheduled period during which planned work is expected to cause outages, so alerts are suppressed and the downtime is labelled planned. The maintenance windows and suppression series covers the mechanics.

Acknowledgement

A person's record that they have seen an alert and are handling it, which stops reminders and escalation. The helpdesk should be able to acknowledge without being able to reconfigure.

Digest

A batch of non-urgent notifications delivered together on a schedule instead of one at a time. Where warnings should go.

Discovery

The process of finding the devices and interfaces on a network, by scanning address ranges and by listening for LLDP and CDP announcements. The discovery and inventory series covers running it continuously.

Inventory and asset management

The list of what you have, where it is and what it is for, kept current by discovery rather than by hand. The discovery and inventory series covers keeping it honest.

Network map and topology map

A drawing of the devices and the links between them; auto-generated from LLDP and CDP, or hand-drawn for the logical layer. The topology types guide and the network documentation series cover both.

Dashboard and wall display

A screen summarising the state of the estate for a person (a dashboard) or for a room (a NOC wall). The dashboard design series covers the difference.

Report, scorecard and SLA report

Periodic summaries for people who do not watch the console: health, availability against a service level agreement, capacity, inventory, alerts. The monitoring data for management series covers what to put in them.

SLA (service level agreement)

A promised level of availability or performance, with consequences for missing it. An SLA report is only worth its paper if the availability figure is coverage-honest. The internal SLAs series covers writing one you can keep.

Retention

How long measurements, events and flow records are kept, and at what resolution; older data is often kept as hourly or daily summaries. The retention and granularity series covers the trade-offs.

Capacity planning

Using the history of utilization and growth to decide when a link, a disk or a server will run out, ideally as a date. The capacity planning from history series covers it.

Synthetic monitoring (active monitoring)

Testing a service by acting as a user on a schedule: pinging, fetching a page, logging in and checking the result. The opposite of passive monitoring, which observes real traffic. The active versus passive monitoring series covers blending the two.

Probe (remote probe)

A component of the monitoring system placed in a branch, a DMZ or a segmented network to poll from the inside and report back, so that no inbound firewall rule is needed and a sensor can be checked from more than one place.

High availability (of the monitor)

Running the monitoring server as a cluster so that the monitor itself surviving a failure is not an act of faith. The monitoring the monitor series covers who watches the watcher.

Self-monitoring

The monitoring server watching its own health: collection failures, poll timing, disk, database latency, and whether it was running at all. A stopped monitor should be a hole in the availability record, never a month of 100%.

Coverage-honest reporting

The rule that an availability or traffic figure is bounded by the time the monitor actually observed, so that a month with no data shows as no data rather than as perfect uptime. The monitoring anti-patterns series lists what happens without it.

Runbook and post-incident review

The written steps for handling a known alert, and the blameless write-up after a real incident. The runbooks and post-incident reviews series cover both.

Monitoring as code

Keeping the monitoring configuration in version-controlled files and applying it through an API, so that changes are reviewed and reversible. The monitoring as code series covers the practice.

Tool sprawl

Running several overlapping monitoring products, each watching part of the estate under its own alert model and login. The monitoring tool sprawl series covers consolidating.


Devices, services and infrastructure


Core, distribution and access switches

The three tiers of a tree-shaped network: access switches in the closets where devices plug in, distribution switches aggregating them, a core joining everything. A failure's blast radius grows with each tier up. The switch port monitoring series covers what to watch on each.

Router and WAN link

The device that forwards between networks, and the link to the outside world or the other site. The router and WAN monitoring series covers watching both, and the ISP accountability series covers proving whose fault the WAN is.

Firewall

The device that decides which traffic passes between zones; monitored for session counts, throughput, CPU, VPN tunnels and the log lines that say what it denied. The firewall monitoring series covers it.

VPN tunnel

An encrypted connection between sites or from a remote worker, monitored for being up, for latency and loss inside it, and for the authentication failures that lock people out. The VPN monitoring series covers it.

Wi-Fi access point and controller

The radios users actually connect through and the system managing them; monitored for client counts, channel utilization, retries and the APs that quietly went offline. The Wi-Fi monitoring series covers it.

VLAN and subnet

A logical network segment on shared switches, and the address range that lives in it. Monitoring them means knowing which devices should be in each and noticing the ones that are not.

Server, service and process

The machine, the thing it provides (a web site, a database, a file share) and the program that provides it. A server can be up while its service is down; monitoring must watch all three.

Hypervisor and virtual machine

The host that runs many virtual servers (VMware ESXi, Hyper-V, Proxmox) and the servers themselves; monitored through the hypervisor's own API as well as through each guest. The hypervisor networking series covers the virtual network between them.

UPS, PDU and environmental sensors

The battery backup, the power strip and the temperature and humidity sensors in a server room, most of which speak SNMP through a network management card. The power and environment monitoring series covers watching the physical world.

Printer monitoring

Reading toner, paper and error state from network printers over SNMP, which every network printer supports and nobody turns on until the one everyone blames runs out at 4 p.m. The printer monitoring series covers it.

Certificate (TLS) monitoring

Watching the expiry date, the protocol version and the cipher suite of every TLS certificate you depend on, so that expiry is a ticket with lead time rather than an outage. The certificate monitoring series covers it.

Website and web application monitoring

Fetching a page or walking a multi-step journey (log in, search, check out) on a schedule and alerting on status, latency and content. The website uptime monitoring and synthetic transactions series cover it.

Database monitoring

Watching a database server's health and, more usefully, running a query whose result is itself the metric: queue depth, replication lag, order backlog.

Mail flow monitoring

Sending a message through the mail system and timing its arrival, so that a stuck queue is detected before the "did you get my email?" calls. The email flow monitoring series covers it.

Cloud monitoring (and hybrid)

Reading metrics from a cloud provider's monitoring API (CloudWatch, Azure Monitor, Google Cloud Monitoring) into the same console as the on-premises estate, so a hybrid estate has one alert model. The cloud network monitoring series covers it.

OT, SCADA and the plant floor

Operational technology: controllers, sensors and building systems that speak Modbus, MQTT or OPC UA rather than SNMP. The IoT and building systems series covers monitoring them with the same tool as the office network.

Out-of-band management

Reaching a device through a path that does not depend on the production network: a management port, a console server, a cellular modem. The out-of-band management monitoring series covers watching the back door.

Rogue device and shadow IT

Something on the network that nobody authorised: an unknown switch, a personal access point, a cloud service the finance team signed up for. The rogue device detection and shadow IT discovery series cover finding them.

IDS, IPS and SIEM

Intrusion detection and prevention systems watch traffic for attack patterns; a security information and event management system collects and correlates logs across the estate. Adjacent to monitoring rather than part of it; the IDS and IPS for generalists and SIEM-lite series cover how much of each a network team needs.

Home lab and home network monitoring

The same discipline at the scale of one router, one switch and a few servers, which is where many administrators learn it. The home lab monitoring practice series covers doing it properly without a budget.


Latency issues (meaning, and the usual causes)

A latency issue is a round-trip time that is higher than the baseline or than the application can tolerate. The usual causes, in order of likelihood: a saturated link somewhere on the path, a device whose CPU is busy, a Wi-Fi hop, a long route, and a DNS resolver that answers slowly so the connection cannot even start.


Define latency in networking

The time a packet takes to travel from sender to receiver, or there and back, in milliseconds. In networking the number that matters is the round trip, because that is what every request waits for.


Define bandwidth: in networking, in a computer network, and internet bandwidth

In networking, the capacity of a link in bits per second. In a computer network it is set by the slowest link on the path, usually the WAN. Internet bandwidth is the rate your provider sells, downstream and upstream. To explain bandwidth in one sentence: the width of the pipe, not the speed of the water.


Define jitter in networking

The variation in latency between consecutive packets, in milliseconds. Real-time traffic cares about it more than about latency itself.


Define uptime

Either the time since a device last started, or the share of a period during which a service was available. Say which you mean; reports that mix them are how 100% gets printed for a month nobody watched.


Define MIB, and define SNMP

A MIB is the catalogue of objects an SNMP device can report, as a tree described in MIB files. SNMP is the protocol that reads those objects. The what is SNMP guide defines both at length.


Community string in SNMP: its purpose

The shared password of SNMP v1 and v2c, sent in clear text with every request. Its purpose is to say which manager may read (or write) a device's objects; its weakness is that anyone who captures a packet has it, which is why SNMP v3 exists.


What is considered high latency, and what p99 latency means

Relative to the baseline, anything the users notice; in absolute terms, above about 150 ms a voice call feels wrong and above a few hundred milliseconds interactive applications crawl. p99 latency is the value that 99% of measurements fall under: the experience of the unluckiest one in a hundred packets, which is usually the one someone complains about.


Packet loss detector (and what 100% packet loss means)

Anything that measures loss continuously: an ICMP sensor, a packet train between probes, or the interface discard counters read over SNMP. 100% packet loss means nothing is getting through at all, which is an outage, not a quality problem.


NetFlow header and record format

A NetFlow packet carries a header (version, record count, uptime, timestamps, sequence number, source ID) followed by flow records; in v9 and IPFIX the records are laid out by templates the exporter sends. The NetFlow analyzer page covers what the records contain.


Availability monitor (its primary function) and network discovery (its primary purpose)

An availability monitor's primary function is to say whether something was reachable and working, continuously, and to keep the record. Network discovery's primary purpose is to find what is actually on the network so the inventory and the monitoring are complete, as opposed to what the spreadsheet says.


Difference between latency and throughput, and between bandwidth and throughput

Latency is how long one packet takes; throughput is how many bytes per second a transfer actually achieves; bandwidth is the ceiling on throughput. High latency lowers throughput on any transfer that waits for acknowledgements, which is why a fast link across an ocean still feels slow.


Where to go next


If a term here raised more questions than it answered, the linked series is the place to take them. New to the subject altogether? Start with the network monitoring fundamentals series, then the core network metrics, and come back here when a vendor's data sheet uses a word you do not recognise. The full library is on the topics page.


Bandwidth of a channel (what bandwidth represents for a connection)

The capacity of one path, in bits per second: what the channel can carry when nothing else is in the way. For a connection between two hosts the bandwidth is the smallest capacity along the path, which is why a gigabit LAN feels slow when the WAN link in the middle is 100 Mbps. Utilization is how much of that capacity is in use; the two are confused constantly.


Put this into practice with SysPeep Monitor

Every term in this glossary is something SysPeep measures, receives or reports from one Windows service: SNMP, WMI, SSH and IPMI polling, NetFlow, sFlow and IPFIX analysis, syslog and trap reception, discovery and mapping from LLDP and CDP, dependency-aware alerting and coverage-honest uptime. Up to 5 devices are free at full depth.

Download Free TrialLearn About SysPeep