Link Aggregation

Link aggregation combines multiple physical Ethernet links into a logical interface called a link aggregation group, or LAG. Switches and routers can use this interface for VLAN transport and Layer 3 connectivity while distributing frames across its member links. The technology is standardized in IEEE 802.1AX, with older documentation often referring to IEEE 802.3ad. Link Aggregation Control Protocol, or LACP, coordinates compatible members and detects whether they can participate in the aggregate.

Aggregation increases total available bandwidth across multiple conversations; it does not normally multiply the bandwidth available to one conversation. Devices typically select a member using a hash of packet fields, such as source and destination addresses and transport ports. Keeping a flow on one member avoids packet reordering. For example, two 10 Gbps links provide up to 20 Gbps of aggregate capacity before overhead, but a single ordinary TCP flow is generally limited to one 10 Gbps member. Several large flows can also hash onto the same member, leaving capacity unused elsewhere.

In hybrid cloud networking, aggregation is configured between the customer network device and a provider's supported physical interconnect endpoint. AWS Direct Connect supports LAGs for eligible dedicated connections using LACP. Members must have the same bandwidth and terminate on the same AWS Direct Connect endpoint at the same location. A LAG does not combine arbitrary connections across different locations or AWS devices. Provider-specific port eligibility and member limits must be checked before ordering circuits.

Physical compatibility remains essential. Confirm the handoff speed, supported optics, fiber type, connector requirements, and optical power levels for every circuit. Configure consistent member settings and verify that LACP places each intended link into forwarding operation. With LACP, at least one side must initiate negotiation in active mode; two passive endpoints do not establish an aggregate. A static bundle configured at one end does not substitute for matching LACP configuration at the other.

VLAN tagging and maximum transmission unit settings are separate from aggregation. For an AWS Direct Connect LAG, virtual interfaces use VLAN tags over the logical connection; adding a member does not create another virtual interface or BGP session automatically. Ensure the intended VLANs are carried consistently and that member links support the required frame size. Jumbo frames require support across the complete traffic path, including the virtual interface type and downstream networks. Successful small pings do not prove that larger packets can pass without fragmentation or drops.

Member failure normally causes affected traffic to move to surviving members, potentially with transient packet loss. Available capacity then decreases. A minimum-active-members threshold can deliberately take the logical interface down when too few links remain, allowing routing to select another path instead of carrying traffic over an undersized bundle. For example, an application requiring more than 10 Gbps may need alternate connectivity when one member of a two-link, 10 Gbps-per-link LAG fails.

A LAG is not equivalent to geographically diverse connectivity. Members can share a router, building, power source, or fiber route, so multiple links may fail together. Use separate interconnect endpoints and locations when those failures must be tolerated, and use routing policy or equal-cost multipath where appropriate between independent Layer 3 paths. During acceptance testing, check individual member counters, LACP state, aggregate throughput with multiple flows, large-packet delivery, and routing behavior after a member failure.