Home › Topics › Dashboards: Build, Buy, or Both › The Four That Earn Their Place

Designing the Few Dashboards That Earn Their Place


The other guides in this pillar give you the rules: one question per screen, one owner per dashboard, one canonical dashboard per question, fewer panels than you want. They do not give you a starting set. You have a blank monitoring tool, a small network, and an empty grid. Which dashboards should exist at all?


For most small and mid-size networks, four. Not four kinds that multiply into forty; four screens, each with a specific question, a specific audience, and a layout you can sketch on paper in a minute. They cover the room, the site, the service, and the future. A fifth may eventually be justified, after these four are built, watched, and trusted. I have seen the fifth built in the first week.


This guide describes each of the four: what it answers, who reads it, what goes on it, and what does not. It then gives layout patterns that apply to all of them. These cover what belongs top-left and when to use a number versus a graph versus a map. It closes with a build checklist. That way, each dashboard leaves your hands with a question, an audience, a refresh, an owner, and a review date attached.


In this guide

  • Why four
  • Dashboard 1: the overview screen
  • Dashboard 2: the per-site view
  • Dashboard 3: the service view
  • Dashboard 4: the capacity view
  • Layout patterns: numbers, graphs, maps, and the top-left corner
  • The build checklist
  • Build these four well before building a fifth
  • Where to go next

Why four


The four dashboards map onto the four questions a small IT team is actually asked. "Is anything broken?" comes from everyone, all the time. "What is going on at the north office?" comes from whoever is dealing with that office. "Why is the phone system flaky?" comes from the business, about one service at a time. "When do we need a bigger link, a bigger disk, a new switch?" comes from whoever holds the budget, once a quarter. Each of those has a different audience, a different time range, and a different answer shape. That is the condition under which one question per screen says to build a separate dashboard.


Four is also the number a small team can maintain. Each dashboard needs an owner and a monthly review, and rots without them. Four reviews a month is a habit. Fourteen is a chore that stops happening, and the maintenance burden describes what happens next.


Dashboard 1: the overview screen


Question: is anything critical broken right now? Audience: whoever is in the room, and the on-call engineer. Time range: now. Refresh: seconds to a minute.


This is the wall screen, the one on the television. Its discipline is that it shows the state of critical services and sites and nothing else. Not graphs, not trends, not the CPU of anything. Use a grid of large colored tiles, one per thing that would cause an immediate flood of complaints if it failed. Those include the internet link, the firewall, the core switch, each site's WAN connection, and the domain controller. They also include the file server, the phone system, email, and the wireless controller. Green when fine, red when not, grey when unknown. If everything is green, the screen is boring, and boring is the correct state for a wall screen.


The temptation is to add "just one graph" so the screen looks busier. Resist it. A graph on a wall screen is unreadable from across the room and draws the eye away from the tiles. The only non-tile element that earns a place is a small network map whose nodes change color with state. A map with a red node reads at a distance. It shows where the problem is as well as that there is one. Include a count of active alerts if your tool can show it as a single number. "3" in a corner is readable, and it tells the room whether the red tile is the only problem.


Leave off anything that is not critical, anything that is a trend, and anything a manager wants to see (they have their own screen). Also leave off every device whose failure would not be noticed by users within the hour. The overview screen is not an inventory. It is a smoke detector.


Dashboard 2: the per-site view


Question: what is happening at this site right now, and what changed today? Audience: the admin dealing with that site. Time range: last day, with the ability to widen to a week. Refresh: a minute or two.


A site is any location with its own WAN connection: the head office, a branch, a warehouse, a clinic. The per-site view is one dashboard with a site selector, not one dashboard per site. That way, every site is laid out identically and an admin who knows one knows them all. Its rows follow the path a packet takes into the building.


The WAN row comes first. It shows the state and utilization of the site's internet or private link, and its latency and packet loss to a known target. It also shows the state of the backup link if there is one. The core row comes next: the site's main switch or router, its CPU, its uplink ports, and the error counters on those uplinks. The key servers row shows whatever local servers the site depends on — a domain controller, a file server, a print server. It shows up/down state, disk headroom, and memory. The PoE row is the one people forget. If the site's switches power access points, phones, and cameras over Ethernet, show the PoE budget. That is the total power the switch can supply versus the power currently drawn. A switch at the top of its budget silently refuses to power the next device plugged in. That failure looks like "the new phone is dead" rather than anything network-shaped. A single gauge per switch stops that ticket before it is written.


What to leave off: every access port. A per-site view with a tile for each of forty-eight ports is a catalog, not a dashboard. Uplinks and the ports feeding servers are the ones that matter; the rest belong in the tool's device page, one click away.


Dashboard 3: the service view


Question: is this business service healthy, and if not, which link in its chain is the problem? Audience: the admin troubleshooting it, and often the manager who owns the service. Time range: last few hours. Refresh: a minute.


A service here means something the business names, not something IT names: "the phone system," "email," "the point-of-sale," "the patient records application." Each one depends on a chain of components, and when the service is flaky the question is always "which link?" The service view draws that chain, left to right, in the order a user's request travels, with the state of each link. The links are the user's site WAN, the firewall, the core switch, the server, the database behind it, and the external provider if there is one. The service view is one dashboard per service, and most small networks have between two and five services important enough to deserve one.


Under the chain go the two or three metrics that predict trouble for this service specifically. For the phone system that is latency and jitter on the WAN and the call server's CPU. For email it is queue length and the state of the mail relay. For the point-of-sale it is response time to the payment provider and the state of the store's link. The choice of metrics is what makes the service view different from a per-site view of the same boxes. The same core switch appears on both, but here it is one link in a chain whose meaning is "phones." The metrics around it are chosen for phones. Working out what belongs in each chain is the subject of dependency thinking; the service view is where that thinking becomes a picture.


The service view is also the dashboard most likely to justify a flexible dashboard tool. A complete chain often wants data from outside the monitoring system: the application's own error log, the helpdesk's open tickets about the service. If it stays single-source, keep it built-in. If it needs those extras, this is where a second system earns its keep, as discussed in built-in versus flexible dashboards.


Dashboard 4: the capacity view


Question: what is going to run out, and when? Audience: whoever plans purchases, and the admin who has to make the case. Time range: the last three to twelve months. Refresh: daily.


The other three dashboards are about now or today. The capacity view is the only one about the future. And its panels all have the same shape: a long, slow trend line for something with a ceiling, and the ceiling drawn on the graph. WAN utilization at the busy hour, trending toward the link's capacity. Disk usage on the file server and the backup target, trending toward full. Uplink utilization on the core switch. DHCP scope usage, trending toward exhaustion. PoE budget across all switches. Wireless client counts per access point. Each graph answers the question "at this rate, when does this hit the line?" — and that is a purchase decision with a date on it.


Two design points make this dashboard useful. First, use the right view of utilization. A graph of raw five-minute samples over six months is noise. A daily busy-hour value or a weekly percentile shows the trend cleanly. The choices are laid out in bandwidth graph views, and they apply just as well to disks and PoE. Second, draw the ceiling and the lead time. If ordering a circuit upgrade takes a couple of months, draw a second line at the utilization where you must start ordering. In that case, do not just draw the line where the link is full. Turning that trend into a purchase date is explained in from utilization to purchase timing.


What to leave off: anything without a ceiling. CPU load is a health metric, not a capacity metric, unless you are planning a server replacement. Latency has no ceiling. Alert counts have no ceiling. If a panel cannot answer "when does this run out?", it belongs on the morning view of the per-site dashboard, not here.


Tip: the capacity view is the one dashboard your manager will thank you for. It turns "we need a bigger link" from an opinion into a graph with a date on it. Build it early, even with only a few weeks of data. The trend line gets more convincing every month, and the day you need it is not the day to start collecting.


Layout patterns: numbers, graphs, maps, and the top-left corner


The four dashboards differ in content but share a few layout decisions. Getting these right is most of what gets a screen read.


What goes top-left

The top-left panel is the one everyone reads first, so it carries the dashboard's single most important answer. On the overview screen, that is the internet link tile, or a "critical alerts" count, because if that is red nothing else matters. On the per-site view, it is the site's WAN state and utilization, because every other row depends on it. On the service view, it is the overall service state, one big tile summarizing the whole chain. That way, the reader knows the verdict before reading the evidence. On the capacity view, it is whichever resource is closest to its ceiling. That means the top-left panel of this dashboard may change from quarter to quarter. Nothing decorative ever goes top-left: not a logo, not a title panel, not a clock.


Number, graph, or map?

Each panel type answers a different kind of question, and choosing the wrong one is the most common layout mistake. A number (a single large value, or a colored state tile) answers "what is it right now?" and is readable at a distance. Use it for state, for counts, and for anything where the reader needs the current value and nothing else. A graph answers "where is it going?" and needs the reader to be close. Use it wherever the trend matters more than the value, which is most of the per-site view and all of the capacity view. A map answers "where is the problem?" and is worth its space only when location is part of the question. That is true on the overview screen and often on the service view, where the dependency chain is itself a kind of map. Maps drawn by hand rot the moment a device is added. The ones worth putting on a dashboard are generated from what the monitoring tool discovers, as described in auto-generated network maps.


A quick check is to count. An overview screen should be mostly numbers. A per-site view is a mix, numbers in the WAN row and graphs below. A service view is one map-like chain plus a few graphs. A capacity view is all graphs, each with a line across it. If the proportions look wrong, the dashboard is probably answering a question other than the one on its sticky note.


Rows follow the path, columns follow the site

A dashboard may have a natural order — the path from the internet into the building, the chain from the user to the database. In that case, lay the panels out in that order, so that reading the screen is the same as tracing the path. Where a dashboard compares like with like — several sites, several disks — put them in columns with identical panels. That way, the odd one out is visible as a difference in shape. Consistency across sites and services is worth more than any individual panel's polish. An admin who has learned one site's screen should read every other site's without thinking.


The build checklist


Every dashboard leaves your hands with seven things attached. Fill this table in before building, not after, because filling it in is what catches the screen with no question, no audience, or no owner. The columns are the four dashboards from this guide, as a worked example for a small network with a head office and two branches.


ItemOverview screenPer-site viewService view (phones)Capacity view
Question answeredIs anything critical broken now?What is happening at this site today?Are phones healthy, and which link is at fault?What runs out, and when?
AudienceThe room; on-callThe admin working the siteAdmin; phone service ownerBudget holder; admin
Data sourcesMonitoring tool stateMonitoring tool: SNMP, ping, WMIMonitoring tool; call server log if availableMonitoring tool history
PanelsNine state tiles, one map, one alert countWAN row, core row, servers row, PoE rowChain of six links; latency, jitter, call-server CPUSix trend graphs with ceilings and lead-time lines
Refresh / rangeThirty seconds / nowOne minute / last dayOne minute / last four hoursDaily / last six months
Owner(name)(name)(name)(name)
Review dateMonthly, first MondayMonthly, first MondayMonthly, first MondayQuarterly, before budget review

With the table filled in, the build itself is short.


  • 1. Name it by audience, purpose, and owner, using the convention from the sprawl guide, so it is findable from day one.
  • 2. Set the time range and refresh at the dashboard level, once, so that every panel inherits them and none disagree.
  • 3. Place the top-left panel first and build outward from it in order of importance.
  • 4. Add the panels from the table and no others. Every extra panel makes the rest smaller.
  • 5. Set the "no data" state of every panel to something that is visibly not green, and put a timestamp on single values.
  • 6. Draw thresholds and ceilings as lines on every graph that has them.
  • 7. Stand back three meters (or ask someone who has never seen it) and check the question is answered at a glance.
  • 8. Add a small text panel with the question, the owner, and the review date. Export the definition to a file next to your monitoring configuration.

Build these four well before building a fifth


This is the part most likely to be ignored. Once these four exist, every additional dashboard will look like a small, reasonable request. A screen for the wireless network. A screen for the printers. A screen the new manager asked for that is almost the overview but with different tiles. Each is easy to build. Each is one more thing to own, review, and keep true. And each is a candidate for disagreeing with the four that already exist.


At Ashgrove Housing the four became eleven in eighteen months, each one a small, reasonable request. A review deleted seven on a Friday, with an archive and a note to the team. Nobody asked where any of them had gone until the following quarter. The seven had been built for meetings. The four had been built for questions.


So before building a fifth, ask three questions. Can the four answer this? A wireless question usually belongs on the per-site view as a row, not on its own screen. A "cameras" question is a service view if cameras are a business service and a per-site row if they are not. Has each of the four been reviewed at least twice and found true? If the existing screens have rot in them, a fifth adds to the rot rather than the value. And would anyone notice if the fifth vanished a month later? If you cannot say who, the request is for a report or an alert, not a dashboard. The guide to what a dashboard is for will help you tell which.


A team that runs four well-kept dashboards for a year will know more about its network than a team with forty. When the fifth is finally justified, it will be because a real question with a real audience appeared that the four could not answer. That is the only good reason a dashboard has ever had.


Where to go next


Four dashboards cover most small networks. An overview screen shows only critical state. A per-site view follows the path from the WAN inward. A service view draws one business service as a dependency chain. And a capacity view turns slow trends into purchase dates. Each has a natural top-left panel and a natural mix of numbers, graphs, and maps. Each also has a checklist entry naming its question, audience, owner, and review date. Build them, keep them true, and leave the rest of the grid empty. The fifth can wait; it always has. The other guides in Dashboards: Build, Buy, or Both cover where those four should live, how to maintain them, and what to do when their numbers start to multiply.


Put this into practice with SysPeep Monitor

If you would rather start from something than from a blank grid, SysPeep Monitor builds live dashboards and network maps. It draws them from the devices it discovers and polls over SNMP, WMI, and SSH, with reports alongside in the same web console. The free trial installs on a Windows machine. An afternoon with it is a practical way to sketch these four screens against your own network before committing to a layout.

Download Free TrialLearn About SysPeep