Reading Lease Churn: What DHCP Logs Say About Your Network
Somewhere on your DHCP server sits the most underrated dataset in your entire network: the lease log. Every device that has ever joined had to stop at the front desk and ask for an address first. That means every laptop, phone, printer, thermostat, and mystery gadget. The desk wrote it down. Time, hardware address, device name, which network, how long it stayed. It is an attendance register for everything on your network, kept automatically, with perfect patience, and read by almost nobody.
That is a waste, because the questions this log answers are the questions admins actually get asked. How many devices do we really have? Is the network growing, and how fast? What joined last night at 2 AM? Why does the guest pool keep filling up? Is something misbehaving? You can buy tools that answer these questions with dashboards and price tags — or you can spend fifteen minutes a week with data you already own.
This guide teaches you to read that register. You will learn what normal lease churn looks like (so abnormal jumps out). You will learn which log entries matter and what each one means. You will learn how to count your true device population and how to turn churn into growth trends and inventory. You will learn the specific anomaly patterns — a device cycling leases, a mass renewal stampede, a flood of strangers — that are worth catching early.
In this guide
- What lease churn is, and what normal looks like
- The log lines that matter
- Counting your real device population
- From churn to growth trends
- Anomalies: the patterns worth catching
- What to graph
- A log-review habit that takes 15 minutes
- Where to go next
What lease churn is, and what normal looks like
Lease churn is the ongoing turnover in your DHCP lease table: new leases granted, existing ones renewed, old ones expiring and returning to the pool. (If leases, scopes, and the four-step DORA handshake are fuzzy, take ten minutes with how DHCP really works first — this guide builds directly on it.) Churn is not a problem to eliminate; it is the network breathing. The skill is knowing your network's normal breathing rhythm, because every interesting discovery in this guide amounts to noticing a change in it.
A typical office network breathes like this. Overnight: near silence, just the occasional renewal from servers, printers, and whatever stays plugged in. Early morning: a swelling wave of activity as people arrive. Some devices are politely renewing leases they still hold. Others (expired overnight or new to the building) are running the full Discover-Offer-Request-Acknowledge exchange. Midday: a gentler hum, with a lunchtime blip on the guest network as visitors arrive and phones hop networks. Evening: activity fades; leases begin quietly expiring. Weekends: near-flat. Layer on top of that a weekly rhythm (busy Mondays, sleepy Fridays if people work remotely) and the occasional planned spike (all-hands day, training class). Then you have your baseline.
Lease duration is the metronome behind all of this, which is why the same network can breathe fast on one scope and slow on another. A guest scope with one-hour leases churns constantly by design — grants and expiries all day, a lease table that empties itself overnight. A wired scope with week-long leases barely murmurs: devices renew midweek and little else happens. Neither pattern is better; they are tuned for different populations (the reasoning lives in the lease-tuning section of the scope exhaustion guide). The practical point for log reading: judge each scope against its own rhythm, not against its neighbors'.
One proportion is worth internalizing: on a healthy, stable network, renewals should heavily outnumber fresh Discovers. Renewals are the sound of a settled population going about its day. A rising share of full DORA exchanges means either the population is churning faster (more transient devices) or something keeps knocking devices off. Neither is automatically bad — but a shift in the ratio is your cue to ask why.
The log lines that matter
Every DHCP server keeps some form of activity log. Windows writes daily audit files, open-source servers log through the system logger, and appliances vary. But they all record the same underlying events, because the protocol only has so many moves. Here is a lightly tidied excerpt of the style you will meet, with the story it tells:
08:41:02 DISCOVER from aa:bb:cc:11:22:33 via 10.0.20.1
08:41:02 OFFER 10.0.20.117 to aa:bb:cc:11:22:33 (LAPTOP-MARIA)
08:41:03 REQUEST 10.0.20.117 from aa:bb:cc:11:22:33
08:41:03 ACK 10.0.20.117 to aa:bb:cc:11:22:33, lease 86400s
09:02:44 REQUEST 10.0.20.63 from 44:55:66:aa:bb:cc (PRINTER-3F)
09:02:44 ACK 10.0.20.63 <-- renewal: REQUEST + ACK, no DISCOVER
09:15:10 REQUEST 10.0.90.24 from de:ad:be:ef:01:02
09:15:10 NAK 10.0.90.24 <-- "not valid here" - device moved networks
11:26:31 DECLINE 10.0.20.88 from aa:bb:cc:77:88:99
<-- client found the address already in use: conflict!
Walk through it once and the grammar sticks. The first four lines are a complete DORA. A laptop arrived with nothing and got 10.0.20.117 for a day (86,400 seconds), via the relay at 10.0.20.1. The "via" tells you which network it was standing on. The printer at 09:02 shows the renewal signature: REQUEST and ACK with no DISCOVER, a settled resident extending its stay. The 09:15 pair is a NAK. A device asked to keep an address from some other network (classic laptop-came-back-from-home behavior). It was refused and would have restarted DORA a moment later. That is harmless in ones and twos, meaningful in bursts. The DECLINE at 11:26 is the rarest and most urgent word in the vocabulary. The client checked its shiny new address, found some other device already squatting on it, and handed it back. That almost always means a static device is parked inside your pool range — a misconfiguration worth chasing the day you see it.
Two other verbs complete the set. One is RELEASE (a device gracefully gave its address back, usually at shutdown). Then there are lease expirations, which some servers log and some just perform silently. That is the whole language — eight-ish words. Everything in the rest of this guide is pattern recognition on top of them. One logistical note: logs on the server disappear whenever they rotate or the box is rebuilt, so forward them to a central collector if you can. The guide to parsing device logs covers how and why in depth.
Counting your real device population
Ask three people how many devices are on the network and you get three guesses, usually all low. The lease data gives you an honest answer two different ways. The lease table (the server's current ledger) answers "who holds an address right now". That is a snapshot, best read at a busy mid-morning moment for your true concurrent peak. The logs answer "who came through over time": count distinct hardware addresses seen in a day, a week, a month. And you have your visiting population, which is always larger and often startlingly so. Both numbers are useful; be precise about which one you are quoting, because they can differ by half.
The counting itself is mundane — extract the hardware-address field, dedupe, count, whether in a spreadsheet or a five-line script. A worked example makes it concrete. The first three bytes of a hardware address are the OUI (Organizationally Unique Identifier), a prefix registered to one manufacturer. Free lookup lists exist for all of them. So a log full of strangers becomes a quick classification exercise. That prefix belongs to a phone maker, and this one to a printer company. That one belongs to a maker of cheap home routers (a prefix that should raise an eyebrow on a corporate network). And the ones marked locally administered are randomized addresses telling you nothing. Ten minutes of prefix lookups routinely turns "forty unknown devices" into "the new phones, two smart TVs, and one thing worth walking over to look at." What matters is what you do with it. Fold the result into your device inventory. Leases carry a hardware address (whose first half identifies the manufacturer), usually a device hostname, the network it appeared on, and timestamps. That is enough to turn "some device" into "a printer, probably third floor, first seen in March." That enrichment work is exactly the subject of building a device inventory. Better yet, run the comparison in both directions. Devices in the leases but missing from inventory are unknowns worth identifying. The systematic hunt is covered in finding unmanaged devices. Meanwhile, inventoried devices that stopped appearing in leases are your quiet decommissions and your lost laptops.
Tip: modern phones and laptops often randomize their hardware address per network to protect privacy, and some rotate it periodically even on the same network. On staff networks where devices reconnect constantly, this can inflate unique-address counts and make one phone look like several strangers. Treat unique-address counts on wireless networks as an upper bound, not a headcount. And expect randomized addresses (many have a telltale digit pattern marking them "locally administered") to dominate guest network logs.
From churn to growth trends
A single count is trivia; a counted series is capacity planning. Take ten minutes at the start of each month to record three numbers per scope: peak concurrent leases, distinct devices seen that month, and scope utilization at peak. Plot them. Within a few months you have the chart that answers the budget questions nobody can currently answer. It shows how fast the device population grows, which network segments are approaching their limits, and when each pool will hit the wall at the current slope. That gives you the luxury of fixing it a quarter early instead of during a scope exhaustion incident. That is what the same discovery looks like when it is made late.
The trend line also catches the growth nobody announces. New devices rarely arrive with a memo; they arrive as a dozen smart displays here, a pilot batch of sensors there, a team quietly doubling its hardware. None of it is loud enough to notice in the moment, and all of it shows in a monthly line that bends upward. When the line bends and you do not know why, the logs from that period will tell you exactly which devices appeared and when. That is more than most organizations can say about their own networks.
Anomalies: the patterns worth catching
With a feel for normal, the abnormal patterns become easy to name. Four are worth memorizing.
One device cycling leases
A single hardware address running full DORA every few minutes — sometimes collecting a different address each time — is a device in trouble. That might mean a flapping wireless connection, a failing network card or cable, a half-broken driver, or something power-cycling in a loop. That cycling wastes addresses, spams your logs, and usually ends with a user complaint days later. The log finds it first. The variant that deserves same-day attention is a device whose grants keep ending in DECLINE. It keeps being handed addresses that something else already occupies. That points at a static-versus-pool overlap in that range rather than at the poor client that keeps reporting it.
A sudden mass renewal or re-join stampede
Dozens or hundreds of devices all running DORA within a few minutes means they all lost their network state together. A switch or access point rebooted, a power blip hit a floor, or your DHCP server itself was down long enough for clients to panic. Remember from the basics guide: clients broadcast for any server when renewals go unanswered too long. So a wave of broadcast renewals is a confession that the server was unreachable. Match the timestamp against your other logs and you have the blast radius and the cause of an outage nobody reported cleanly.
A flood of unknown devices
A burst of never-before-seen hardware addresses has a short list of explanations. Benign: an event, a training class, a visiting audit team. Structural: someone bridged a network they should not have. That might be a personal access point or misconfigured lab box leaking another network's population into yours. That is the same species of mischief as a rogue DHCP server, and worth the same hunt. Hostile but rare: a deliberate flood of forged requests intended to drain the pool — effectively a denial-of-service against DHCP. The response to all three starts identically: where are these addresses appearing (which scope, which relay), and do the manufacturer prefixes look like phones, lab gear, or randomized noise?
The flatline
The subtlest anomaly is silence: a scope whose activity simply stops. No grants, no renewals logged via that network's relay, nothing. Churn never truly stops on a live network segment — something always renews eventually. So a flatline usually means the requests are no longer reaching the server. That might mean a broken or misconfigured relay, a routing change, or a VLAN that quietly lost its helper setting. Users on that segment will not notice until their leases start expiring, which gives you a head start measured in hours if you are watching. This is exactly the failure a per-network synthetic lease test is designed to catch, and it is covered in monitoring DHCP itself.
A NAK spike
A steady trickle of NAKs is normal life — devices wander between networks and get corrected. A spike means a population of devices is asking for addresses that are no longer valid. That is classic after a VLAN renumbering, a subnet change, or a restored-from-backup server with a stale scope. The spike is also a known symptom of two DHCP servers disagreeing about who owns a range. If a change went in yesterday and NAKs jumped today, the log just graded your change.
What to graph
You do not need a data science platform; you need a handful of counters on a dashboard next to your other graphs. If you track only what is in this table, you will catch everything described above.
| Metric | What it tells you | Worth attention when |
|---|---|---|
| Leases granted per hour (new DORA) | Arrival rate of devices needing addresses | Spikes outside the daily rhythm; sustained climb |
| Renewals per hour | The settled population's heartbeat | Mass surge (shared outage) or a broadcast-renewal wave (server was unreachable) |
| Ratio of renewals to new grants | Population stability | Ratio falls — churn is rising and deserves a why |
| NAKs per hour | Devices asking for invalid addresses | Any spike, especially after a network change |
| DECLINEs | Address conflicts detected by clients | Any occurrence at all — investigate same day |
| Distinct new hardware addresses per day | Strangers arriving | Floods; also a slow climb nobody planned |
| Scope utilization per scope | How full each pool is | Warn near 80%, act by 90–95% |
The last row is the headline metric of this whole pillar and has its own guide. The others are cheap to derive from logs with modest scripting or a log-collection tool that can count matching lines per interval. Graphs beat raw thresholds here because churn is so rhythmic. The same absolute number can be normal at 9 AM and alarming at 3 AM. A human glancing at a weekly graph catches that instantly.
A log-review habit that takes 15 minutes
None of this requires heroics — it requires a recurring calendar slot and a short ritual. Weekly works for most networks:
- 1. Glance at the graphs. Utilization per scope, grants per hour, NAK count. Anything off-rhythm from last week?
- 2. Search for the loud words. Any DECLINEs? Any "no addresses available" style complaints from the server? Both are act-now findings.
- 3. Skim the strangers. List the week's first-seen hardware addresses; eyeball the manufacturer prefixes. Expected phones and laptops, or a batch of surprises?
- 4. Spot-check one busy scope's lease table. Does the count roughly match the number of humans and devices you believe exist there?
- 5. Note the numbers down. Peak leases, distinct devices, utilization — three numbers per scope into the trend sheet. Future-you will be grateful.
Keep a few months of log history if storage allows — enough to compare "this Monday" with "a normal Monday" and to investigate something a user reports weeks late. And when you make a network change, drop a one-line annotation in your notes. Half the value of historical graphs is being able to line a bend in the curve up against the change that caused it.
Where to go next
Lease churn analysis is the observational half of DHCP monitoring: it tells you what the service has been quietly recording about everyone else. The operational half is watching the service itself: scope fill levels, failover health, and the synthetic tests that catch a dead server before the morning rush. That operational work is covered step by step in monitoring DHCP itself. For the foundation under all of it, how DHCP really works explains the protocol these logs are narrating. The rest of the series lives at DHCP: The Service Everyone Forgets.
Put this into practice with SysPeep Monitor
Reading the logs tells you who came through the front desk; a monitoring tool tells you what they are and how they are doing. SysPeep Monitor auto-discovers the devices on your network and its interfaces. It collects health metrics over SNMP, WMI, and SSH, and puts the graphs and alert thresholds this guide keeps mentioning on one web console. That is a natural home for your scope utilization and device-count trends. The free trial is a quick install on any Windows machine.
Download Free TrialLearn About SysPeep