Duplicate IP Address Explained: Two Devices, One Address
An IP address is supposed to name exactly one device. When two devices claim the same one, nothing fails cleanly — the service works, breaks, then works again, because traffic goes to whichever device answered last. This guide explains how the clash happens, how devices detect it with ARP probes and announcements, what Cisco's %IP-4-DUPADDR message is really telling you, and the ordered hunt that finds the second device: read the message, watch the ARP table, trace the MAC address to a switch port.
The short answer
What is a duplicate IP address? It is two devices on the same network both claiming the same IP address as their own. An IP address is meant to work like a house address — one address, one building. When two devices use it, the network has no way to know which one a packet was meant for, so deliveries go to whichever device most recently announced itself. The result is the most confusing symptom in networking: a service that works, then breaks, then works again, with nothing obviously down. Routers and switches that spot the clash log it — on Cisco gear the message is %IP-4-DUPADDR — and that message, read properly, tells you the disputed address and the hardware address of the other device claiming it.
Think of two shops on the same street both putting up the same house number. The post still gets delivered — just not reliably to the right shop. A parcel for the first shop arrives at the second because the second was the last one the driver saw with that number on the door. Both shops insist, honestly, that the address is theirs. That is a duplicate IP: the address system itself (called ARP, explained below) keeps a notebook mapping the address to a hardware identity, and the two devices keep overwriting each other's entry in everyone's notebook.
Why a duplicate breaks things in such a strange pattern
On a local network, a sender never delivers a packet to an IP address directly. It looks the IP address up in its ARP cache to get a MAC address, and delivers to that. With one rightful owner, that mapping is stable for hours. With two claimants, the mapping becomes a tug of war: each device announces itself — when it boots, when it is asked, when its own timers fire — and every listener updates its cache to the announcement it heard last. So a server can be reachable for twenty minutes, vanish for five, and return, purely because the other device happened to speak. Sessions drop mid-transfer because the second half of a conversation is delivered to a device that never saw the first half. And both devices look healthy when you test them individually, because each one genuinely owns a working network card and a plausible configuration — the fault is in the combination, not in either box.
One boundary matters: this is a same-subnet disease. ARP only operates on the local network segment, so two devices can only fight over an address if they share a subnet (the same VLAN, in switched terms). The same address used on two different subnets is not a duplicate — it is just reuse, and routers keep those worlds apart. When a log message names a VLAN, it is telling you which world the fight is in.
Neither device is broken. The fault lives in the shared notebook — every listener's ARP cache — being rewritten by two authors.
How devices detect it: probes, announcements, and defence
The good news is that a well-behaved device checks before it moves in. The procedure is standardised in RFC 5227, IPv4 Address Conflict Detection, and it has three acts. Probing: before using an address, the device broadcasts a small number of ARP probes — ARP requests whose sender IP address is set to 0.0.0.0, precisely so the probe itself does not pollute anyone's ARP cache. The probe asks, in effect, "is anyone already using the address I want?" If anyone answers, or if a probe from another device wanting the same address is overheard, the address is taken and the device must not use it. Announcing: if the probes come back silent, the device announces itself — broadcasts ARP announcements in which both the sender and target address are its new address (the same packet shape people call a gratuitous ARP) — so every listener replaces any stale cache entry with the fresh, correct one. Ongoing detection and defence: detection does not stop at boot. For as long as the device uses the address it keeps watching for a conflicting announcement from a different hardware address, and it may defend its address by announcing again — within limits, so two stubborn devices do not flood the network arguing forever.
That machinery explains the messages you actually see. A desktop operating system that loses the argument pops up an "IP address conflict" warning and may stop using the address. A Cisco router or switch that owns an address and hears another device claim it does not give the address up — it logs the clash and keeps a record of who the other claimant is. Which brings us to the message that brings most engineers here.
What %IP-4-DUPADDR actually tells you
The decoder entry for %IP-4-DUPADDR walks the message itself; its sample reads Duplicate address 192.168.10.50 on Vlan10, sourced by mac 001a.2b3c.4d5e. Every field is working information, not decoration. The address (192.168.10.50) is the prize being fought over. The interface (Vlan10) is the subnet where the fight is happening — and, read carefully, it is an address this device itself owns or serves on that interface, which is how it noticed. Sourced by names the MAC address of the other device — the one that sent the ARP traffic claiming the address. That MAC is the single most valuable string in the message: MAC addresses are traceable through a switched network in a way IP addresses are not, because every switch keeps a table of which MAC lives on which port.
Two cautions before you hunt. First, one message can be a passing event — a device briefly bridging two networks, a failover, a laptop docking — so confirm the clash is live (the checks below) before you start unplugging things. Second, some shared addresses are arranged: first-hop redundancy protocols deliberately let two routers serve one virtual gateway address in a coordinated way (see HSRP vs VRRP vs GLBP explained), and RFC 5227 itself carves out rare cases arranged by prior coordination. A duplicate is an uncoordinated claim — two devices that do not know about each other. If the "second device" turns out to be a redundancy partner answering for a virtual address, you are looking at a different fault, not this one.
The hunt: finding the second device, in order
Step 1 — prove the flip. On the router or switch for that subnet, run show arp | include 192.168.10.50 (use the address from your own message) a few times, minutes apart. One stable MAC address means one current claimant. The MAC changing between checks — or differing from the MAC you know belongs to the rightful owner — proves two devices are answering. The decoder's first check is exactly this, because everything after it depends on knowing both MACs: the owner's and the impostor's.
Step 2 — turn the impostor's MAC into a port. On the switch attached to that subnet, run show mac address-table address 001a.2b3c.4d5e (again, your message's MAC, not the sample's). The table answers with the port that MAC was last learned on. If that port is an uplink to another switch, walk: repeat the same command on the next switch, and the next, until the port is an access port with one device on it. This walk is the whole trick — it converts a log line on a router into a physical socket in a physical room, using nothing but tables the switches already keep.
Step 3 — name what is on the port. Port descriptions, CDP or LLDP neighbour output on that port (show cdp neighbors / show lldp neighbors), the DHCP server's lease list, and the first three bytes of the MAC (the vendor prefix) usually identify the device between them: a printer that kept an old address, a server rebuilt last week, a virtual machine that was cloned and powered on beside its original, a test box someone racked "for five minutes". Note what it is before you change anything — you are about to renumber somebody's device, and you want to know whose.
Step 4 — free the address, then verify. Give one of the two devices a different, genuinely free address (or remove the impostor from the network, if it should never have been there). Then re-run step 1: the ARP entry should sit on one MAC and stay there, the log should go quiet, and the flaky service should stay up through the period when it used to flap. If the entry still flips, there is a third claimant or you renumbered the wrong twin — go back to step 2 with the new sourced-by MAC.
The four usual causes — all of them bookkeeping
The site's decoder entry lists the causes engineers actually find, and none of them is an attacker. A static address typed inside the DHCP pool: someone hand-configures a device with a tidy-looking address that DHCP later hands to somebody else — two honest devices, one address, because the pool and the sticky notes disagreed. An old static address surviving a renumber: the network moves to a new range, everything is updated except one server or printer, which wakes up still claiming an address that has since been given a new owner. A cloned virtual machine: a VM is copied as a backup or a test, both copies are powered on, and both boot with the same static address — the clone is identical to the original in exactly the way that hurts. A foreign box with a famous address: a home router, lab device, or loaner plugged into the office network still configured with a common default like .1 or .254 — which, on many networks, is the gateway's or a server's address. The shared shape is worth noticing: in every case the record of who owns which address was incomplete, out of date, or ignored. Duplicates are a documentation fault that expresses itself as a network fault.
Prevention: make the record keep itself
Three habits prevent nearly all of it. Keep static addresses outside the DHCP pool, on purpose: carve the pool and the static range apart (for example, DHCP hands out the middle of the subnet; infrastructure statics live below it, printers and reservations above it — the exact split matters less than having one), and treat typing an address from inside the pool as the fault it is. Prefer reservations to hand-typed statics: a DHCP reservation pins an address to one MAC in one central list, so the "record" and the mechanism are the same thing — a device that must keep its address gets it from DHCP every time, and the pool can never hand it to anyone else. Write clones and renumbers as procedures, not favours: a cloned VM gets a new address before it joins the production network, and a renumber checklist ends with a sweep — ping and ARP-scan the old range and confirm nothing still answers — because the device you forgot is the one that will announce itself at 2 a.m. Larger networks add switch-level guards on top: DHCP snooping builds a binding table of which MAC legitimately holds which address, and Dynamic ARP Inspection checks ARP traffic against that table, so a device claiming an address it was never given finds its announcements dropped at the port. Those features are seatbelts; the three habits are not crashing.
One neighbouring protocol note, so you do not go looking for this message where it cannot appear: %IP-4-DUPADDR is an IPv4 story, because ARP is an IPv4 mechanism. IPv6 does the same job differently — a device checks a new address with Duplicate Address Detection, sending Neighbour Solicitations for it before it may use it — so the symptom exists there too, but the messages and the tables you hunt through are the IPv6 ones.
For the address system underneath all of this, read what happens when you ping across VLANs; for planning address ranges that do not collide in the first place, read the subnetting cheat sheet.
Try it yourself
Draw the subnet the fight is happening on with the free Config-to-Topology Mapper — paste the configs and see which devices attach where before you start tracing MACs — and check a two-device configuration for exactly this class of mistake with the free Network Config Verifier, which flags duplicate addresses in a pasted config set before they ever reach the wire. And when the log line itself is what you have, open the decoder entry for %IP-4-DUPADDR and work its checks in order — ARP table first, MAC table second, address change last.
Fact-check sources: the probe / announce / ongoing-detection procedure, the ARP probe's 0.0.0.0 sender address, the announcement's sender-equals-target shape, address defence, and the carve-out for arrangements made by prior coordination — RFC 5227, IPv4 Address Conflict Detection (which updates RFC 826, the original ARP specification). The %IP-4-DUPADDR sample message, its meaning, its four listed causes, and its two checks (show arp for the address; show mac address-table for the sourced-by MAC) are quoted from the site's own decoder entry, which is sourced from Cisco IOS IP addressing documentation. IPv6 Duplicate Address Detection via Neighbour Solicitation is the mechanism described for IPv6 address configuration in RFC 4862. DHCP snooping's binding table and Dynamic ARP Inspection's use of it to validate ARP are Cisco switching security features as described in Cisco's Catalyst security configuration documentation.
Related guides
- A duplicate IP address means two devices on the same subnet are answering for one address. Senders cannot tell them apart, so traffic lands on whichever device answered the last ARP request — which is why the symptom is intermittent, not a clean outage.
- Devices police this themselves: before using an address, a host probes for it with ARP requests whose sender address is 0.0.0.0, then announces itself; detection continues for as long as the address is in use (RFC 5227). A Cisco router or switch that hears another device claim its address logs %IP-4-DUPADDR and names the other device's MAC address.
- The log message is the map: it gives you the disputed address, the interface or VLAN, and the MAC address of the other claimant. The hunt is then mechanical — confirm the ARP entry flips between two MACs, then follow the impostor MAC through the MAC address tables, switch by switch, to one access port.
- The usual causes are bookkeeping, not malice: a static address typed inside the DHCP pool, an old static address left behind after a renumber, a cloned virtual machine powered on beside its original, or a test box or home router plugged in carrying a common address like .1 or .254. The fix is one free address and a record that stays kept.