What is network interface bonding failover detection?
If you are running mission-critical hardware, a single broken cable or a failed switch port shouldn't take down your entire server. Network Interface Bonding (also known as Teaming or LAG) allows you to combine multiple physical NICs into one logical interface.
However, bonding is only as good as its Failover Detection. If the hardware doesn't know a link is dead, it will keep sending packets into a "black hole."
There are two primary ways hardware and software work together to detect a failure: MII Monitoring and ARP Monitoring.
This is the most common and basic form of detection. It looks at the "Link State" of the NIC.
How it works: It asks the hardware driver, "Is the green light on?" If the cable is unplugged or the switch loses power, the MII status drops to DOWN, and the bond immediately moves traffic to the healthy link.
The Flaw: MII only detects physical failure. If the cable is fine, but the switch has a software hang or a "zombie" port that isn't passing traffic, MII will still think the link is UP.
This is a much more robust, "logical" check.
How it works: The bond sends ARP requests to a known IP address (like your gateway) at regular intervals. If it doesn't get a response back on a specific link, it assumes that path is dead—even if the physical light is still green.
The Benefit: It detects "Silent Failures" and upstream issues that MII might miss.
When using LACP (Link Aggregation Control Protocol), the failover detection becomes even more intelligent. LACP uses LACPDUs (Control Protocol Data Units)—small heartbeats sent between the server and the switch.
Fast vs. Slow: You can configure these heartbeats to be "Slow" (every 30 seconds) or "Fast" (every 1 second).
The Benefit: If the switch stops responding to these heartbeats, the server knows within seconds that the link is unreliable and pulls it out of the active "bundle."
When configuring failover on Oracle hardware or Linux servers, these three settings determine how "twitchy" or "stable" your network is:
| Parameter | What it Does | Why it Matters |
| miimon | How often (in ms) to check the link state. | Too low can cause "flapping"; too high makes failover slow. |
| updelay | How long to wait before bringing a "repaired" link back into the bond. | Prevents a flaky cable from rapidly connecting and disconnecting. |
| downdelay | How long to wait before declaring a link dead. | Helps ignore tiny "blips" in the hardware signal. |
Standard High Availability: MII Monitoring is usually sufficient for 90% of use cases. It is simple, fast, and has zero network overhead.
High-Reliability/Cloud: Use LACP with Fast Heartbeats. This ensures that both the server and the switch agree on the health of the link.
Complex Topologies: If your server is connected to a "dumb" switch that doesn't support LACP, ARP Monitoring is your best bet for detecting upstream failures.
Failover detection is the "brain" of network redundancy. Without it, you don't have a backup—you just have a second point of failure. By properly tuning your monitoring intervals, you can ensure that your hardware reacts to a crisis in milliseconds, before your users even realize there's a problem.