Home › Topics › Monitoring Licensing Models and Hidden Costs › Growth Traps

Growth Traps: How Licensing Bites When the Network Grows


A monitoring license is sized for the network you have on the day you sign. And the network stops being that size about a week later. A branch opens. A project adds a rack of virtual machines. Someone swaps a small switch for a big one, and the internet link gets an upgrade. Every one of those is good news for the business and a silent event for the license. That lasts right up until the console shows a warning that you are over your count. Or the renewal quote arrives with a number nobody budgeted for. I have had both happen in the same month. The warning was the polite one.


None of this is a trick. Vendors publish how their licenses count, and the counting is honest. The trap is that growth happens in units you were not watching. You were counting devices while the license was counting interfaces. You were counting sites while the license was counting probes. This guide names the seven traps that catch small and mid-size teams and explains how each one bites. It shows you how to plan headroom. That makes growth something you buy on purpose rather than something that is announced to you.


If you have not yet read the companion guide on what each licensing model counts, start with Licensing Models in Plain Language. This guide assumes you know the difference between a device unit, a sensor unit, and a flow unit.


In this guide

  • Why growth is where licensing hurts
  • The seven growth traps, one at a time
  • The polling angle: more sensors also means more load
  • Planning headroom from your own growth rate
  • The growth-trap table
  • A worked example in abstract units

Why growth is where licensing hurts


Licenses are step functions. A step function is a line that stays flat and then jumps. Nothing changes for a long time. Then one small event moves you to a new level all at once. Most costs in IT are smooth, so people are trained to expect that a small change costs a small amount. Licensing does not work that way. Adding the fifty-first device to a fifty-device license is not a fifty-first-of-a-license event; it is a whole-new-band event.


The second reason growth hurts is that it arrives sideways. The network grows in the dimension the business cares about (more users, more sites, more capacity). Meanwhile, the license grows in the dimension the vendor counts (more interfaces, more flow records, more probes). The two are related but not equal, and the gap between them is where surprises live. (Nobody in a planning meeting has ever said "we are adding four hundred interfaces".) Each of the traps below is a case where a normal growth event turns into a larger-than-expected licensing event. The cure is the same for all of them: track the license's unit rather than the business's unit. And look at growth one renewal ahead instead of one quarter behind.


Trap 1: Tier thresholds


How it bites. Many licenses are sold in bands: up to one count, then up to a larger count, then larger again. The band is a threshold. One more device than the band allows moves you to the next band. In some contracts the next band reprices the entire entitlement, not just the extra device. A single small addition can therefore change the whole renewal line. Bands also interact with editions. Sometimes the larger band is only available in a higher edition. In that case, crossing a size line also crosses a feature line.


How to plan. Know where your next threshold is, in the license's own units. Write it on the same page as your current count, so the two numbers are never read apart. When you are within a normal quarter's growth of the line, decide deliberately. Trim what you monitor, or move up before the renewal so the step is budgeted. Ask, in writing, whether crossing a band reprices the whole entitlement or only the increment. The answer changes how much headroom is worth buying in advance.


Trap 2: Sensor counts that explode when you add a switch


How it bites. Under per-sensor licensing, a switch is not one unit; every monitored interface is, and often every metric on every interface is. The arithmetic is multiplication, and multiplication is where junior admins get ambushed. One 48-port access switch, monitored with a default template that watches status and traffic on every port, looks like this:


48 ports x 2 sensors per port (status, traffic)   =  96 sensors
+ device sensors (ping, CPU, memory, temperature,
  two power supplies)                             =   6 sensors
--------------------------------------------------------------
one switch, default template                      = 102 sensors

Replace a small switch with a stack of four of these, and you have consumed several hundred units with one change ticket. Add error counters and discards per port and the number rises again. Nobody bought the switch intending to spend that many units; the template did it.


How to plan. Treat the monitoring template, not the switch, as the thing you are budgeting. Decide which interfaces earn a sensor: uplinks, ports feeding servers and access points, and ports with a history of trouble. Access ports to desks usually do not need traffic sensors; port-status changes there are noise. A curated template for the same switch might be two uplinks with status and traffic, four server ports with status, and the six device sensors. That is sixteen units instead of a hundred. Build that template before the switch arrives, and make "which template?" a line in the change request.


Trap 3: Flow volumes that grow with bandwidth


How it bites. Flow licensing that counts records per second or stored volume does not care about your device count at all. It cares about traffic. When the internet link is upgraded, when a new cloud application becomes the office's daily tool, when video meetings replace phone calls, or when a scanner runs amok, the record rate rises. Some collectors handle overage by sampling (keeping one record in every N and estimating the rest), which keeps the picture but blurs the detail. Others drop records or stop accepting them. Then the "top talkers" report quietly becomes fiction during the busiest hour, which is exactly when you wanted it.


How to plan. Watch your record rate the way you watch disk space: as a trend, not a snapshot. Note the rate before and after any link upgrade. If your license is per exporter rather than per rate, the trap is instead the number of branch routers you enable flow on. In that case, enable it deliberately, starting with the core and the internet edge. Ask the vendor what happens at the cap. And prefer behavior you can see (a visible sampling rate, an alert) over behavior you cannot (silent drops).


Trap 4: Virtualization sprawl


How it bites. Under per-device licensing, every virtual machine with its own address is a device. Virtualization makes devices nearly free to create. A template and a right-click produce a new VM in minutes. A test VM that "will be deleted next week" becomes a permanent resident; a project asks for twenty desktops. Each one is a unit. The hypervisor hosts (the physical machines the VMs run on) are units too. And a container platform can add a whole new class of things that may or may not count, depending on the definition. The device count moves without anyone touching a cable.


How to plan. Decide which VMs deserve monitoring at the device level and which are covered adequately by monitoring the host and the platform. Temporary and lab VMs can live in an unmonitored group, or be monitored by ping only under a model where that is cheap. Put a "monitoring?" question in the VM request form. I went two years without one, and a license warning was how I learned there was a desktop pilot. If your VM count grows faster than your physical count, this is the shape where an unlimited or site license deserves a serious look.


Trap 5: Remote probes that each need a license


How it bites. A probe (also called a collector or a poller) is a small piece of the monitoring system installed at another site or in another network segment. This lets the central server watch devices it cannot reach directly. Probes are how you monitor branches, the warehouse, the network behind the lab firewall, and the segment on the far side of a VPN. A VPN is a virtual private network, the encrypted tunnel that joins two sites over the internet. Some licenses count each probe as a device; some sell probes as separate units; some include a fixed number and charge for the rest. Every new site therefore carries a probe cost on top of its device and sensor costs. The probe is the item people forget when they estimate "one router, one switch, three access points."


How to plan. Count sites, not just devices, and keep the probe rule from your license next to that count. Ask whether one probe can serve several small sites over the VPN. It often can, at the cost of polling traffic across the tunnel. Also ask whether probes in the same site can share a license. Include the probe in the standard new-site checklist. Some products also cap the number of sensors a single probe may carry. So a probe can hit its own ceiling before the overall license does. Note that limit beside the probe count.


Northgate Retail opened its ninth shop on a Monday, with a router, a switch, three access points and a probe. All of those had been on the project plan since spring. The probe had not been on the license plan. That was because the license plan was a device count and nobody had thought of the probe as a device. The count crossed the band at nine in the morning. The console showed a warning at five past. For the rest of that week the new shop was monitored by its manager ringing head office. The purchase order for the next band went through the following Friday, at the speed purchase orders go. The new-site checklist now has a line that says "probe license" and another underneath that says "check the band". Both were added by the person who spent that week on the phone.


Trap 6: High-availability pairs counted twice


How it bites. Two firewalls in an active-passive pair are one firewall to your users and two devices to your monitoring tool. That is because each has its own management address and its own health. The same is true of clustered switches, paired load balancers, and redundant hypervisor hosts. Under per-device licensing, redundancy doubles the count for those devices, and under per-sensor licensing it doubles their sensors. The monitoring system itself can fall into this trap too. A high-availability pair of monitoring servers may need two server licenses. The guide on high-availability monitoring covers that alongside the question of whether the pair earns its cost at all.


How to plan. Count every management address, not every logical device. When you tally redundant pairs, write both members down and ask whether the license offers any allowance for a passive member. Some do not, and that is fair; you are monitoring two boxes.


Trap 7: Auto-discovery that adds devices you did not mean to license


How it bites. Auto-discovery is a gift for inventory and a hazard for licensing. A scan of the subnet finds the printers, the phones, the smart TVs, the badge readers, the vending machine with an Ethernet port, and the visitor's laptop. A tool configured to "add everything found" adds them, consuming units for devices you never intended to monitor. Under per-sensor licensing the effect is larger, because discovery also finds every interface on every device and can attach sensors to all of them. The count jumps overnight, and the first sign is a license warning. The second sign is an uptime graph for the vending machine.


How to plan. Separate discovering from licensing. Let discovery find everything (that is its job for the inventory), but require an approval step before a found device becomes a monitored, licensed device. Keep a "found, not monitored" group where new devices wait. The guide on keeping the inventory current shows how a scheduled discovery with a review step keeps the inventory fresh. The same review step is where you decide what gets a license.


The polling angle: more sensors also means more load


There is a second cost to every sensor that no quote shows. Each sensor is polled. Polling consumes CPU on the monitoring server, bandwidth on the WAN (the wide area network: the links between your sites), and processing time on the monitored device. A licensing decision to add a hundred sensors is therefore also a capacity decision. This is one of the few places where the license and the engineering agree: fewer, better-chosen sensors are cheaper in units and cheaper in load. The arithmetic of how many sensors a server and a WAN link can comfortably poll is worked out in the guide on the polling load budget. Keep that budget beside the license count, because you will hit one of the two limits first. It is useful to know which.


Planning headroom from your own growth rate


Headroom is the distance between your current count and the license limit. Too little, and every project becomes a licensing conversation. Too much, and you are paying for capacity you will never reach. The right amount is not a rule of thumb; it is a calculation from your own history. And you already have the data, in a device list that nobody has ever sorted by date.


  • Measure your growth rate from the inventory. Use your inventory (or the monitoring tool's own device list, with its "added on" dates). It tells you how many devices, interfaces, and VMs joined in each of the last few quarters. Average it. That is your baseline growth in each unit the license cares about.
  • Add the known projects. Growth is not only drift. A planned branch, a desktop virtualization rollout, a switch refresh, or a link upgrade each has a known unit impact. List them with the quarter they land in.
  • Project one renewal ahead. Current count, plus baseline growth for the term, plus project impacts, equals the count you will have at the next renewal. Buy the band you will be in then, not the band you are in now.
  • Keep a small reserve. Leave enough spare units for one unplanned quarter. That way, an emergency replacement or an urgent project does not force a mid-term purchase at a bad moment.
  • Re-check quarterly. Compare the projection with the actual count each quarter. If you are ahead of the line, the trims above (templates, VM groups, discovery approval) buy time until the renewal.

Tip: the cheapest headroom is the headroom you create by monitoring less of the wrong things. Before you buy a band up, run a quick review of what consumes units today. Look for unmonitored-in-practice test VMs, desk ports with traffic sensors, and retired devices still in the tool. Reclaiming those units is free and usually embarrassing in a useful way.


The growth-trap table


Growth trapHow it bitesHow to plan
Tier thresholdsOne extra unit moves you to the next band, sometimes repricing everythingKnow the next line; ask whether crossing reprices all or the increment; buy next term's band
Sensor multiplicationInterfaces times sensors per interface; one switch is a hundred unitsCurated templates; sensors only on uplinks, server ports, trouble ports
Flow volumeRecord rate follows bandwidth and behavior, not device countTrend the rate; enable exporters deliberately; know the behavior at the cap
Virtualization sprawlEvery VM is a device; templates make devices free to createDecide which VMs are device-monitored; unmonitored lab group; question on the VM form
Remote probesEach site needs a probe, and probes may be licensed unitsCount sites; keep the probe rule beside the count; probe on the new-site checklist
High-availability pairsRedundant members are counted individuallyCount management addresses, not logical devices; ask about passive-member allowances
Auto-discoveryEverything found becomes licensed, including printers and visitorsSeparate discovery from licensing; approval step; "found, not monitored" group

A worked example in abstract units


Meet a fictional company with one main office and one warehouse. The numbers below are unit counts only; no money appears anywhere, because the point is to see the counting. The license in question is a per-device band that allows up to one hundred devices, sold alongside a per-sensor allowance.


Today

Main office
  firewall pair (active/passive)         2 devices
  core switch stack (3 members)          3 devices  (license counts members)
  access switches (48-port)              8 devices
  wireless access points                12 devices
  hypervisor hosts                       2 devices
  virtual machines                      35 devices
Warehouse
  router + switch                        2 devices
  remote probe                           1 device   (license counts probes)
--------------------------------------------------------
current count                           65 devices   (band: up to 100)

Sixty-five is comfortable inside a band of one hundred, and a new admin would reasonably stop worrying here. The project list in the drawer is why you should not. Now add the plans for the coming term.


The coming term

Baseline drift (from inventory history)
  about 3 devices per quarter x 4        12 devices
New branch office
  router, switch, 3 access points         5 devices
  remote probe                            1 device
Desktop virtualization project
  new virtual desktops                   40 devices
Switch refresh
  replace 8 access switches              0 devices  (like for like)
--------------------------------------------------------
projected count at renewal             123 devices   (band: up to 100 exceeded)

The count crosses the band during the term, and it crosses because of the virtualization project, not because of anything the network team did. The trap has revealed itself in a spreadsheet rather than in a license warning, which is the entire point of doing the projection. There are now three honest options, and the team can choose one calmly. The first is to buy the next band at renewal with the project in the budget. The second is to keep the virtual desktops out of device monitoring and watch them through the hosts and the desktop platform instead. That holds the count near eighty-three. The third is to move to a model whose unit does not follow VM count at all.


The switch refresh is the quiet part of this example. It adds no devices, but under the per-sensor allowance it is the largest event of the term. Eight new switches with a default template of about a hundred sensors each is eight hundred units. The same eight switches with a curated template of sixteen sensors each is one hundred twenty-eight. The device count says "nothing happened." The sensor count says "everything happened." That difference is why you track both.


Where to go next


Growth traps are not vendor tricks. They are the places where the network's growth and the license's counting disagree. And you now know the seven that matter, how to read your own growth rate from the inventory, and how to buy headroom deliberately. Do the projection once a term and the license warning stops being how you find out. The renewal quote becomes a number you wrote down first. The next guide, The Hidden-Costs Checklist, moves from unit counts to everything else on a quote. Meanwhile, Tracking Licenses and Usage turns this projection into a quarterly habit. Both live in Monitoring Licensing Models and Hidden Costs.


Put this into practice with SysPeep Monitor

The projection in this guide is only as good as the inventory it starts from. SysPeep Monitor discovers devices and interfaces automatically and shows what it is monitoring from one web console. That makes counting your current units a report rather than a walk around the building. Pricing is on the site, and a free trial lets you run the count on your own network.

Download Free TrialLearn About SysPeep