Maps, Weathermaps, Heatmaps, Tables: Choosing the Right Picture
Every monitoring tool ships with a map, and the first thing most people do is put everything on it. I did. Two hundred devices, every link, a photograph of the region as a background, and a legend with nine colours. It was beautiful for a week and then it was wallpaper, because nobody could answer a single question by looking at it.
The problem was not the map. The problem was that I had not decided what question the map was for. A picture of the network is a tool for answering one question quickly. Each kind of picture is good at one question and bad at the rest. That includes status maps, weathermaps, heatmaps, line graphs, and plain tables. Choose the picture for the question and it works. Choose the picture because it looks impressive and it will answer the question badly, or, worse, answer a different question with great confidence.
By the end of this guide you will be able to name the question each visual answers. You will be able to say what each one hides, and pick one from a decision table. You will also be able to run a two-minute test that tells you whether your picture can be read by someone who has never seen it. There is also a short story about a geographic map, which I include because I have built its cousin.
In this guide
- A visual answers one question
- The five candidates
- What each one hides
- The decision table
- The wall versus the browser
- The new-starter test
- A war story: six sites, one building
- The choosing checklist
- Where to go next
A visual answers one question
A network visual is a picture that changes as the network changes. It earns its place by answering one question faster than the raw data could. "What is down right now?" "Which link is busy?" "Which of these ninety access points is behaving oddly?" Those are three different questions, and the picture that answers one of them well answers the others badly.
The word "faster" is doing all the work in that sentence. Your monitoring system already holds the answer to every one of those questions in its database. A visual is not new information. It is a layout that lets your eye find the answer in a second instead of your fingers finding it in a minute. If a picture does not save that minute, it is decoration. The sister pillar on dashboard design has a great deal to say about decoration.
So the first step in choosing a visual is to write down the question. One sentence, with a question mark at the end. If you cannot write it, you are not ready to choose. If the sentence has an "and" in it ("what is down and how busy is it?"), you need two pictures, not one bigger one. (The bigger one is always suggested. It is never the answer.)
This guide is about live visuals: pictures that repaint themselves every time the monitoring system polls. A poll is one round of the monitoring system asking every device how it is, usually every one to five minutes. The static drawing you keep for documentation, with cable labels and rack positions, is a different object with different rules. And the guide to network diagrams covers it. A documentation diagram is true the day you draw it. A live visual is true only while the polling behind it is.
The five candidates
There are five kinds of picture worth knowing, and most monitoring tools can produce all five. This section defines each one and names the question it answers. The guides that follow in this pillar go deep on the first three; here the job is only to tell them apart.
The status map: what is down, and where
A status map is a drawing of your devices and the links between them, with each one coloured by its current state. The colours are green for up, red for down, and, if the map is honest, grey for "we do not know." It answers "what is down right now, and where in the network is it?" Its strength is location. A red box in the branch-office corner tells you which building to worry about before you have read a single device name. Its weakness is that it shows nothing about how busy anything is, or how long it has been that way. Building a live status map covers layout, colours, and the dependency colouring that stops one dead uplink from painting forty devices red.
The weathermap: which links are busy
A weathermap is a topology map where the links, not the devices, carry the colour. And the colour means utilisation: how much of a link's capacity is in use right now, as a percentage. The name comes from the road-traffic maps on the evening news, green roads and red roads. It answers "where is the network busy?" across the whole estate in one glance. It does not answer "is this device up?" (the boxes are usually plain) and it does not answer "was it busy an hour ago?" The weathermap guide covers the colour scale, showing direction, and the trap of drawing a link at the wrong speed.
The heatmap: which of many things is unusual, or when
A heatmap is a grid of coloured cells, where each cell's colour is a value. Rows might be devices and columns hours of the day; then one glance shows you which device is unusual and at what time. It answers the question no map can: "out of a hundred things, which ones are different?" A status map with a hundred boxes is a crowd. A heatmap with a hundred rows is a pattern. The heatmap guide covers the grids that matter and a short script that builds one from a spreadsheet export.
The line graph: how one thing changed
A line graph is time along the bottom and one value up the side. It answers "how did this one thing change, and when?" It is the best picture ever invented for a single metric over time and a poor one for anything else. Five lines on one graph is a conversation. Twenty is a rug. If your question is about one link, one CPU, one latency figure, the line graph wins. None of the maps in this pillar are trying to replace it.
The table: exact values, sortable
A table is rows and columns of numbers with a header you can click to sort. It answers "what are the exact values, and which is the largest?" It is the least glamorous picture on this list and it wins more often than people admit. When you need the top ten interfaces by utilisation, a table sorted by utilisation gives you the answer. In that case, it gives you the exact number and the device name too, all in one row. A weathermap gives you a colour and a guess.
I will commit to an opinion here. If you are unsure which picture to build, build the table first. It takes ten minutes, it is exact, and it will tell you whether the question was worth a picture at all. Half the maps I have been asked for were replaced by a sorted table and nobody missed them. A table has never once asked for a bigger monitor.
A table can be produced from any export with a single command. Suppose your monitoring tool exports the current interface utilisation as a CSV file. CSV means comma-separated values: plain text, one row per line, columns separated by commas. Suppose the file has the columns device, interface, and percent. This one-liner sorts it by the third column, highest first, and shows the top ten:
sort -t, -k3 -nr interfaces.csv | head -10
The same idea in PowerShell, for a Windows admin who would rather not open a second shell:
Import-Csv interfaces.csv | Sort-Object { [double]$_.percent } -Descending | Select-Object -First 10
Neither line is clever. That is the point. If the sorted list answers the question, you have saved an afternoon of dragging boxes around a canvas.
What each one hides
Every picture hides something, and the thing it hides is usually the thing that goes wrong next. This is not a flaw to fix; it is the price of answering one question well. The discipline is to know what your picture hides and to have another picture, or a table, that shows it.
| Visual | Shows well | Hides |
|---|---|---|
| Status map | Current state and location | Load, history, and any device that was never drawn on it |
| Weathermap | Current load per link, across the estate | Device state, history, bursts shorter than one poll, and a link drawn at the wrong speed |
| Heatmap | Patterns across many things over time | Exact values, the cause, and anything between the time buckets |
| Line graph | One value over time | Everything else, plus scale tricks (an axis that starts at 90% turns a ripple into a cliff) |
| Table | Exact values, sorted | Location, relationships, and trend |
Two of those deserve a second look. The status map's blind spot is the worst, because it is silent. A device that was added to the network and never to the map is not red, not grey, just absent. And absence looks exactly like health. Keeping maps truthful is the guide for that failure and its relatives.
The table's blind spot is relationship. A table can tell you that three links are at ninety percent. It cannot show you that all three hang off the same distribution switch, which is the fact that turns three alerts into one problem. That is the moment a map earns its keep. It is the reason the answer to "map or table?" is usually "table, then map if the table keeps making you scroll."
The decision table
Below is the table I keep in the team wiki. It maps a question to a picture, names what that picture hides, and says how often it should refresh. Copy it, then add a row every time somebody asks for "a map" and you make them say the question out loud first.
| Question | Best visual | What it hides | Refresh rate |
|---|---|---|---|
| What is down right now, and where? | Status map with dependency colouring | Load, history, unmapped devices | Every poll (1–5 min), with a stale marker |
| Which link is busy right now? | Weathermap | Bursts shorter than a poll, device state | Every poll (1–5 min) |
| Is this one link, CPU, or latency figure getting worse? | Line graph | Everything else | On demand, at the poll interval |
| Which of many devices is unusual, and when? | Heatmap, device by time | Exact values, the cause | Hourly or daily |
| What does latency look like over a day, a week, a month? | Heatmap, value by time | Individual samples | Hourly |
| Which ports on this switch stack are busy or erroring? | Port heatmap | Direction, history | Every poll |
| What are the exact values, sorted? | Table | Location, relationships | Every poll or on demand |
| What is the network's shape, for documentation? | Static diagram (not a live visual) | Current state | On change, by hand |
The refresh-rate column is there because a picture that refreshes faster than the data behind it is lying about its own freshness. If the monitoring system polls every five minutes, a map that repaints every ten seconds is repainting the same numbers thirty times. Match the refresh to the poll and put the poll interval on the picture, so the reader knows how old "now" is. Choosing polling intervals is the guide for deciding the poll itself.
The last row is on the table on purpose. People ask for a live map when what they need is a drawing they can print. It is kinder to say so than to build a live map that is used as a drawing. Tools that draw the topology for you from discovered data are a third thing again. The guide auto-generated maps covers what they get right and wrong.
The wall versus the browser
The same question needs a different picture depending on where it is read. A wall screen is read by a room from three metres away, in passing. A browser is read by one person, close up, with a mouse. The NOC wall guide covers wall-screen design and visual hierarchy and glanceability covers size, colour, and position. I will not repeat their rules. What matters here is which of the five pictures survives the walk to the wall.
On the wall, only the status map and a very small table survive. A weathermap on a wall is a coloured tangle from the door. A heatmap is a fuzzy rectangle. A line graph is a line. The wall's job is to look different when something breaks, and a status map with large boxes does that. A table of the five worst things does it too. Nothing else on the list does.
In the browser, everything works, and the weathermap and heatmap come into their own, because the reader can hover, click, and read a legend. Build your weathermap and your heatmaps for the browser. If somebody asks for them on the wall, ask what question the room needs to answer from the door. (It is never "which of the ninety access points had a busy Tuesday.")
One picture, two homes, is the mistake to avoid. A status map designed for the browser does not become a wall map by being shown on a bigger screen. It becomes a small map far away. If a question needs answering in both places, build two pictures from the same data and give them different names.
The new-starter test
A picture that needs a legend to be read is a picture that will be misread when it matters. Here is the test I use, and it takes two minutes. Find the newest person on the team, or anyone from outside IT who is willing. Put the picture in front of them with no legend and no explanation. Ask three questions.
- "What is broken right now?" They should point at something, or say "nothing," within five seconds.
- "Where is it?" They should name a site, a building, or a floor from the layout alone.
- "Is this picture current?" They should find a timestamp or a freshness marker without being told where to look.
Each failure points at a specific fix. If they fail the first question, the colours are wrong or there are too many of them. Three states plus grey is the most a stranger can read at speed. If they fail the second, the layout does not match how the organisation thinks about its own network. In that case, the boxes need to be arranged by site or floor rather than by where they happened to land. If they fail the third, the picture has no freshness marker. That is the most dangerous failure of the three. That is because a frozen picture and a healthy network look identical.
Fix the failure and run the test again with a different person, because the first one now knows the answers. The test also cures the habit of designing for yourself. You know what the orange means because you chose the orange. The person reading the picture at two in the morning did not choose anything.
Tip: keep the legend anyway, small and in a corner. The test asks whether the picture can be read without it, not whether it should exist. A legend is for the tenth question, not the first three.
A war story: six sites, one building
Meridian Parts had a handsome geographic status map: a photograph of the region with a pin for each of its six sites. Five of the six "sites" were floors of the same building. So five pins landed on the same pixel and the map showed one pin, mostly green. When the third-floor distribution switch failed, four green pins drawn over one red one produced a shade. In the incident notes, the on-call engineer described it as "sort of khaki." Finding the red pin took twenty minutes and five overlapping tooltips. The map was replaced the following week with six boxes in a column, labelled by floor. And it has been dull and useful ever since.
The lesson is not "never use a geographic background." Geography is the right layout when the question is "which town," and Meridian's question was "which floor." The layout of a picture is part of its answer. And a layout that does not match the way people already think about the network makes them translate under pressure. The status map guide has the layout rules: sites as boxes, links flowing in one direction, no crossings where you can avoid them.
The choosing checklist
Run this list before you build any live picture, and again when somebody asks you to "just add one more thing" to an existing one. Every item is a yes or a no.
- The question is written down in one sentence with a question mark.
- The sentence has no "and" in it; if it did, there are now two questions and two pictures.
- The visual was picked from the decision table, not from what the tool shows by default.
- What the visual hides is written next to it, with the name of the picture or table that shows the hidden thing.
- A table of the same data was built first and was found wanting.
- The picture has one home: the wall or the browser, not both.
- The refresh rate matches the poll interval, and the poll interval is shown on the picture.
- A new starter answered the three test questions without a legend.
- The picture shows a timestamp or freshness marker.
- The picture has a named owner and a review date.
The last item looks like paperwork and is the one that keeps the others true. A picture without an owner drifts. A device is added and not drawn. A link is upgraded and not relabelled. A colour is changed "just for now." Six months later the picture answers a question about a network that no longer exists. Every live visual needs a name on it and a date on which that person will look at it with fresh eyes. Maps breed when nobody owns them, and dashboard sprawl describes the litter.
Where to go next
Question first, picture second. The status map answers "what is down and where." The weathermap answers "where is it busy." The heatmap answers "which of many things is unusual." The line graph answers "how did this one thing change." And the table answers "what are the exact numbers." Each hides something the others show. So most networks end up with a small set of pictures, each with a question written on it, rather than one big map that answers nothing.
The rest of this pillar takes each live picture in turn. It covers building a status map that tells the truth, weathermaps, and heatmaps for time and scale. It also covers the discipline of keeping maps truthful. All of them live under Visualizing Network Health: Maps, Weathermaps, and Heatmaps.
Put this into practice with SysPeep Monitor
The five pictures in this guide are drawn from the same monitoring data. It helps when they live in one place. SysPeep Monitor builds network maps and live dashboards from the devices and interfaces it auto-discovers. It uses the same web console for thresholds, dependencies, and alert rules, so a parent failure does not flood the map with red. The free trial installs on one Windows machine and finds the network for you.
Download Free TrialLearn About SysPeep