Weathermaps: Painting Link Load Onto the Topology
The first weathermap I built was the most popular picture in the department for a month. It showed every link between our sites, coloured by how busy it was. People who had never opened the monitoring console would stop by to look at it. Then a manager used it to argue that a circuit did not need upgrading, because the map said it was quiet. The map was wrong, in a way I will explain, and it took a week of slow file transfers to convince anyone.
A weathermap is the most persuasive picture monitoring can produce, and that is exactly why it needs more care than any other. A status map that is wrong shows a red box where there should be a green one, and somebody notices within the hour. A weathermap that is wrong shows a calm green link that is in fact saturated, and nobody notices until the bill, or the complaints, arrive.
By the end of this guide you will be able to say what a weathermap is for. You will be able to design a colour scale a stranger can read. You will be able to show direction so inbound and outbound are not averaged into a lie. You will be able to tell the map's number from the bill's. You will be able to keep the link speeds true and build one from interface counters. There is a link inventory you can copy and a short story about a link that was eight percent busy.
In this guide
- What a weathermap is
- What it is good for, and what it is not
- Designing the colour scale
- Showing direction
- Now versus the 95th percentile
- Keeping the link speeds correct
- Building one: counters, drawing, refresh
- The link inventory template
- A war story: eight percent busy
- The weathermap checklist
- Where to go next
What a weathermap is
A weathermap is a topology map where each link is coloured by its utilisation. That is the share of a link's capacity in use right now, as a percentage. The devices are drawn as plain boxes and the lines between them carry the information. That is the way a road-traffic map draws towns in grey and roads in green, amber, and red. The name is borrowed from the weather forecast, and it is apt: a weathermap shows conditions, not causes.
The number behind each colour is simple. A link has a capacity, say one gigabit per second, which means it can carry one billion bits every second. The monitoring system reads the link's traffic counter every few minutes. It works out how many bits crossed it per second on average since the last reading. Divide that by the capacity and you have utilisation. Eight hundred million bits per second on a one-gigabit link is eighty percent, and the line goes red.
A weathermap is not a status map with extra colours. A status map answers "what is down and where?"; a weathermap answers "where is the network busy?" Combining them produces a picture where a red line might mean "saturated" or "dead," and the reader has to hover to find out which. Keep them separate; choosing the right picture covers which question needs which map.
What it is good for, and what it is not
A weathermap is good for a capacity glance across the whole network. Twenty links, one screen, one second, and you know where the load is. A table of twenty utilisation figures makes you read twenty numbers; a map makes you see one red line. It is also good for spotting the one hot link among many quiet ones. It is also good for showing management where the network is busy, in a picture they can read without you standing beside it.
A weathermap is not a troubleshooting tool for a single link. When one link is red, the next questions are "since when?", "how often?", and "who is using it?", and the map answers none of them. The first two need a line graph of that link over the last day and week. The third needs flow data, which records which addresses sent the traffic. The map's job is to tell you which link to graph. Once it has done that, close it.
A weathermap is also not a substitute for a graph, and this is the mistake that catches the most people. The map shows one number per link: the latest one. A link that is red on the map might have been red for ten minutes or ten weeks. And those are different problems with different fixes. Uplink saturation covers what a persistently busy uplink looks like and what to do about it; the map only tells you where to look.
Designing the colour scale
Use few bands with boundaries that mean something. Four is enough: quiet, moderate, busy, saturated. A scale with ten bands looks precise and reads as mush, because nobody can tell the sixth shade from the seventh at a glance. The boundaries should be numbers you would act on, not round numbers that look tidy. If seventy percent sustained is where you start planning an upgrade on your links, seventy is a boundary. How full is too full is the guide for choosing those numbers; the map just paints them.
| Band | Utilisation | Colour | Width | Meaning |
|---|---|---|---|---|
| Quiet | 0–50% | Pale green | Thin | Nothing to do |
| Moderate | 50–70% | Dark green | Medium | Watch the trend |
| Busy | 70–85% | Amber | Thick | Plan the upgrade or shaping |
| Saturated | 85–100% | Red | Thickest | Users are feeling it now |
| No data | — | Grey, dashed | Thin | Counter not read; do not assume quiet |
Put a legend on the map, and make the map readable without it. The legend states the band boundaries in numbers, because "amber" means different things on different maps. The map should still work for a reader who never looks at the legend, which is why the table has a width column.
Colour alone is not enough, because a share of your readers cannot tell red from green. A colour-blind-safe map carries the same information in a second channel. That can be line width (thicker is busier), a small percentage label on each link, or both. Width is the better of the two on a wall, where labels are too small to read; the label is better in a browser. A map that only works for readers with typical colour vision is a map that will be misread by someone on the on-call rota.
The grey band is the one people forget, and it is the one that keeps the map honest. A link whose counter has not been read is not quiet; it is unknown. Draw it grey and dashed, never pale green. Keeping maps truthful covers why "no data shown as fine" is the most dangerous default in monitoring.
Showing direction
A link has two utilisations, not one, because traffic in and traffic out are different numbers. A branch office downloading a large update might run at ninety percent inbound and five percent outbound at the same moment. A map that shows one colour per link has to pick one of those numbers or average them. And the average, forty-seven percent, is a value the link never had. (It is a comfortable dark green, though.)
The standard fix is to draw each link as two halves, each coloured by its own direction. One convention is two half-arrows: the half nearest device A is coloured by traffic leaving A, and the half nearest device B by traffic leaving B. Another is two thin parallel lines, one per direction. Either works. What matters is that the reader can see, without hovering, that a link is busy one way and idle the other. That is because that pattern is a diagnosis in itself.
Write down which end is "in." Every interface counter reports traffic from the point of view of the device that owns it. It reports the octets that arrived at that port and the octets that left it. "Inbound on the core's port" is "outbound on the branch's port." When two people build a map from two ends of the same link, they disagree about which half is which unless the convention is written on the link inventory. It always is written down, eventually. Usually after the argument.
Now versus the 95th percentile
The map shows now; the bill and the plan use the 95th percentile, and they are different numbers. The map's colour comes from the last poll: the average bits per second over the most recent interval, usually one to five minutes. The 95th percentile is a summary of a whole month of those intervals. Sort every sample from smallest to largest, throw away the top five percent, and take the largest that is left. Carriers bill on it, and capacity plans should be built on it. That is because it says what the link does on a normal busy afternoon rather than during one spike. The 95th percentile explained walks through the arithmetic.
Do not put the percentile on the map. A weathermap is a live picture, and a live picture painted with a monthly summary is neither live nor a summary. It is a link coloured by what it did last month, refreshing every five minutes for no reason. If someone needs the percentile, give them a table of links with the 95th percentile as a column, sorted. It answers their question and it does not pretend to be weather.
Be honest about what "now" hides. Even a five-minute average smooths over bursts. A link that ran at one hundred percent for thirty seconds and idled for the rest of the interval shows as ten percent. The users who waited during those thirty seconds were not imagining it. A map that averages over an hour hides even more. It can show a link as comfortably moderate while it saturates every few minutes. Granularity blind spots covers what disappears between samples. The map is a good first look at load and a poor last word on it.
Keeping the link speeds correct
A link drawn at the wrong capacity lies in both directions. Utilisation is bits per second divided by capacity. So if the map thinks a link is ten times faster than it is, every colour on that link is ten times too calm. If it thinks the link is slower than it is, the map screams about a link that is fine. The counter is right in both cases. The denominator is wrong, and nothing on the map tells you.
The device's own idea of its speed is a starting point, not the truth. Over SNMP, an interface reports its speed in a value called ifHighSpeed, in megabits per second. That value is what the port negotiated, which is usually right for a port on a switch and often wrong for a WAN circuit. A carrier that sells you two hundred megabits delivered on a one-gigabit handoff will have a port that says one gigabit. Its contract will say two hundred megabits. The contract is the capacity that matters. Set the speed on the map by hand for every link like that, and write down where the number came from.
Speeds change and the map does not know. A ten-gigabit port may have negotiated down to one gigabit after a bad optic. The carrier may have applied a circuit upgrade last Tuesday. A link may have been moved to a different port. Each one changes the denominator, and the map keeps using the old one until somebody tells it. The link inventory below exists so that somebody can.
Tip: once a quarter, compare each link's inventory speed against what the device reports and against what the contract says. When the three disagree, the contract usually wins for a rented circuit and the device usually wins for a switch-to-switch link. Write down which one you chose.
Building one: counters, drawing, refresh
The data source is the interface counters, read over SNMP. SNMP (Simple Network Management Protocol) is the standard way a monitoring system asks a network device about itself. Every business-grade switch and router keeps a running count of the bytes that have entered and left each port. The counters you want are the 64-bit ones, ifHCInOctets and ifHCOutOctets. The older 32-bit counters wrap around to zero after about four gigabytes, which a one-gigabit link can do in under a minute. Interface metrics that matter covers the whole family.
Reading them by hand looks like this, using the standard command-line tools and interface index 5 on a device at 10.0.0.1 with the read community public:
snmpget -v2c -c public 10.0.0.1 IF-MIB::ifHCInOctets.5 IF-MIB::ifHCOutOctets.5 IF-MIB::ifHighSpeed.5
A counter on its own is not a rate. You need two readings and the time between them. Suppose the inbound counter reads 1,200,000,000,000 and, three hundred seconds later, 1,230,000,000,000. The difference is thirty billion octets, which is 240 billion bits (eight bits per octet). Over three hundred seconds that is 800 million bits per second. On a one-gigabit link, that is eighty percent. On a ten-gigabit link, eight. Same counter, same traffic, two different colours, and only the speed decides which.
# utilisation % = (delta_octets * 8) / (seconds * speed_bits_per_second) * 100 (30000000000 * 8) / (300 * 1000000000) * 100 = 80
The drawing is the topology, laid out with the same rules as a status map. Traffic flows top-down or left-to-right, sites are boxes, and links do not cross where a re-route avoids it. Most monitoring tools that draw maps can bind a link's colour to an interface's utilisation directly. If yours cannot, the tool's map usually accepts a value per link from a query. In that case, the query is "latest utilisation for this interface."
The refresh should match the poll and no faster. A map that repaints every ten seconds from a five-minute counter is repainting the same colour thirty times. A reader who sees it flicker will assume it is more current than it is. Put the poll interval in the legend so the reader knows how old "now" is.
The link inventory template
This is the table that keeps a weathermap honest, and it is the one piece of paperwork I would not skip. One row per link on the map, filled in before the link is drawn, and reviewed when the map is reviewed. The example rows belong to the Northgate Retail network from the status map guide.
| Link | A end (device, port) | B end (device, port) | Speed | Counter source | Band thresholds |
|---|---|---|---|---|---|
| DC core to head office | core-a, port 1/49 | dist-1, port 1/49 | 10 Gbit/s (device) | core-a, ifIndex 49, ifHCIn/OutOctets | 50 / 70 / 85 |
| North branch WAN | core-a, port 1/12 | rtr-north, port 0/0 | 200 Mbit/s (contract; port says 1 Gbit/s) | rtr-north, ifIndex 2, ifHCIn/OutOctets | 50 / 70 / 85 |
| South branch WAN | core-b, port 1/12 | rtr-south, port 0/0 | 100 Mbit/s (contract; port says 1 Gbit/s) | rtr-south, ifIndex 2, ifHCIn/OutOctets | 50 / 70 / 85 |
| dist-1 to acc-1 | dist-1, port 1/1 | acc-1, port 1/48 | 1 Gbit/s (device) | dist-1, ifIndex 1, ifHCIn/OutOctets | 50 / 70 / 85 |
Three columns do the work. The speed column records the capacity and, in brackets, where the number came from. So the next person knows whether to trust the device or the contract. The counter source column names the exact device and interface index the colour is read from. So nobody has to guess which end of the link the map is listening to. The thresholds column lets a link have its own bands if it needs them; a backup circuit that is meant to be idle might go amber at twenty percent.
A war story: eight percent busy
Acme's weathermap showed the link between its two main sites as eight percent busy at the height of the working day. Everyone found that reassuring until the finance team's month-end transfers started timing out. The map had been drawn from a template that assumed every core link was ten gigabits. And this one was a one-gigabit circuit that had been provisioned at one gigabit because that was what the budget allowed. The counter was correct. The denominator was ten times too large. The link was running at eighty percent and the map called it eight. The upgrade request that had been declined on the strength of the map was approved the following week. That was on the strength of the same map with one number corrected. The colour of the line changed from pale green to red without a single packet changing.
This is the failure a weathermap is most prone to and the hardest to see. A wrong speed does not look wrong. There is no error, no grey link, no gap in the graph, just a calm colour that happens to be a lie. The only defence is the inventory, and the quarterly check that the inventory still matches the world.
The weathermap checklist
Run this before the map is shown to anyone whose decisions it might inform, and again at every review.
- The map answers "where is the network busy?" and nothing else; device state is on a separate status map.
- The colour scale has four bands plus grey, with boundaries taken from your own utilisation thresholds.
- The legend shows the band boundaries in numbers and the poll interval in minutes.
- Line width or a percentage label carries the same information as the colour.
- A link with no current counter reading is grey and dashed, never green.
- Every link shows both directions, and the inventory says which end is "in."
- Every link's speed is in the inventory with its source: device, contract, or measured.
- Every rented circuit's speed is the contracted rate, not the port's negotiated rate.
- The refresh interval matches the poll interval.
- The 95th percentile is available in a table, not painted on the map.
- The map has an owner and a review date, and the inventory is reviewed on the same date.
Where to go next
A weathermap is a counter, a clock, and a denominator, drawn onto a topology. The counter is almost always right. The clock decides how much the map smooths over. The denominator is where the lies live, and the inventory is how you find them. Keep the bands few, the direction visible, and the speeds written down. And the map will do the one thing it is good at, which is telling you where to look next.
When the question is not "which link" but "which of a hundred things, and when," the map runs out of room and a grid takes over. The article heatmaps for time and scale is the next guide. The one after it, keeping maps truthful, is the monthly discipline that catches the wrong denominator before finance does. Both live under Visualizing Network Health: Maps, Weathermaps, and Heatmaps.
Put this into practice with SysPeep Monitor
A weathermap is only as good as the interface polling underneath it. SysPeep Monitor auto-discovers devices and their interfaces. It collects traffic counters over SNMP from a central Windows service or optional remote probes. It draws network maps and live dashboards in the same web console it uses for thresholds and alert rules. The free trial installs in an afternoon, which is about how long the link inventory takes to fill in.
Download Free TrialLearn About SysPeep