Port Aggregation/LACP, MTU, Jumbo Frames, and Interface Settings
Key Takeaways
- Link aggregation bundles multiple physical links into one logical link for redundancy and load sharing.
- LACP (802.3ad) dynamically negotiates the bundle; static aggregation skips negotiation and is riskier.
- Aggregation hashes each flow onto one member link - a single flow does not span all links.
- Standard Ethernet MTU is 1500 bytes; jumbo frames (~9000 bytes) must be enabled end-to-end.
- Port-channel members must match in speed, duplex, mode, VLANs, native VLAN, MTU, and LACP mode.
Beyond VLANs: Real Interface Settings
Switch implementation is more than assigning VLANs. Production links also need redundancy, matched physical settings, and consistent frame sizes. Network+ tests whether you can spot the one mismatched value that breaks a bundle or a transfer.
Link Aggregation and LACP
Link aggregation combines two or more physical links into one logical link, called a port channel, EtherChannel, LAG (Link Aggregation Group), or bond depending on the platform.
| Concept | Meaning |
|---|---|
| Link aggregation | Many physical links act as one logical link |
| LACP (IEEE 802.3ad) | Standards-based dynamic negotiation of the bundle |
| Static aggregation | Manual bundling, no negotiation - mismatch risk |
| Load sharing | Flows distributed by a hash of header values |
| Redundancy | Bundle survives loss of a member link |
Key exam point: Aggregating four 1 Gbps links gives 4 Gbps of aggregate capacity, but any single TCP flow is hashed onto one member and is still capped near 1 Gbps. Many flows are needed to use the full bundle. Watch for the trap that says "bond two links to double one backup's speed" - that is wrong.
| Port-channel setting | Why it must match across members |
|---|---|
| Speed | Mismatched speed suspends the member |
| Duplex | Mismatch causes errors and slow links |
| Access/trunk mode | Bundle is one logical interface |
| Allowed VLAN list | Trunk members must carry the same VLANs |
| Native VLAN | Untagged traffic must be consistent |
| MTU | Frame-size mismatch drops large frames |
| LACP mode (active/passive) | At least one side must be active |
MTU and Jumbo Frames
MTU (Maximum Transmission Unit) is the largest payload a link will carry in one frame. Standard Ethernet MTU is 1500 bytes. Jumbo frames raise this to about 9000 bytes and are used in storage, backup, and virtualization networks to cut per-frame overhead.
| MTU topic | Exam clue |
|---|---|
| Standard 1500 MTU | General-purpose Ethernet |
| Jumbo ~9000 MTU | Storage (iSCSI), backups, vMotion |
| MTU mismatch | Small pings work, large transfers fail |
| Path MTU | The smallest MTU along the whole path wins |
| DF bit + oversize | Packet dropped instead of fragmented |
Critical rule: jumbo frames must be enabled end to end - every NIC, switch port, uplink, router, and storage interface in the path. Enabling 9000-byte MTU on the server and one switch but leaving the uplink at 1500 produces the classic symptom: a 64-byte ping succeeds, but a large file copy or backup stalls or fails.
Worked example: After a storage team sets MTU 9000 on a host and its access port, ping -l 1472 -f (a 1500-byte frame with DF set) succeeds, while ping -l 8972 -f (a ~9000-byte frame) fails with "packet needs to be fragmented." That gap proves the path drops oversize frames somewhere - usually an uplink still at 1500.
Speed, Duplex, and Autonegotiation
| Setting | Healthy state | Bad clue |
|---|---|---|
| Speed | Negotiates or matches peer | One side forced, other auto |
| Duplex | Full duplex on switched links | Late collisions, slow throughput |
| Autonegotiation | Same policy both ends | One forced, one auto = duplex mismatch |
| Error counters | Low and stable | CRC, runts, giants, drops rising |
| Interface status | up/up | admin down or err-disabled |
A duplex mismatch (one side full, one side half) is the classic cause of a link that works but is painfully slow with late collisions and CRC errors. The fix is to set both sides the same way - usually autonegotiation on both.
Reading Interface Counters
| Counter | Likely cause |
|---|---|
| CRC errors | Bad cable, connector, or transceiver; sometimes duplex mismatch |
| Runts | Frames smaller than 64 bytes - collisions or faults |
| Giants | Frames larger than allowed MTU - jumbo/MTU mismatch |
| Late collisions | Duplex mismatch or cable too long |
| Drops / discards | Congestion, buffer exhaustion, or policy |
PBQ Decision Table
| Requirement | Likely configuration |
|---|---|
| Redundant high-speed uplink | LACP port channel |
| Storage network, large frames | Jumbo MTU on every interface in path |
| One member suspended | Compare speed, duplex, VLANs, MTU, LACP mode |
| Small traffic OK, backups fail | Check MTU/jumbo consistency end to end |
| CRC errors on a port | Inspect cable, transceiver, duplex |
Common Traps
| Trap | Better reasoning |
|---|---|
| One flow uses all bundled links | Flows are hashed onto one member |
| Jumbo frames on only the endpoint | Every device in the path must support it |
| Mix access + trunk members in one bundle | Members need identical switching mode |
| Ignore physical errors during VLAN work | Bad optics/cable mimic higher-layer faults |
| Force speed on one side only | Match autonegotiation policy on both ends |
When a member silently drops out of a port channel, dump the bundle summary and compare it line by line to a healthy member - the suspended link almost always differs in exactly one setting.
Which protocol dynamically negotiates a link aggregation bundle between two devices?
Small pings succeed, but large file transfers across a storage VLAN fail after jumbo frames were enabled on only one switch. What is the most likely issue?
Which settings must be consistent across port-channel member links? Select two.
Select all that apply