Link aggregation refers to the implementation of a trunk link that acts as a direct point-to-point link, between two devices such as peering routers, switches, or arouter and switch combination at each end of the link. The link aggregation comprises of links that are considered members of an Ethernet trunk, and build an association which allows the physical links to operate as a single logical link. The link aggregation feature supports high availability by allowing the physical link of a member interface to switch traffic to another member link in the event that a particular interface fails. In aggregating the links, the bandwidth of a trunk interface is combined, equaling the sum of the bandwidth of all member interfaces,to enable an effective bandwidth increase for traffic over the logical link. Link aggregation can also implement load balancing on a trunk interface. This enables the trunk interface to disperse traffic among its member interfaces, and then transmit the traffic over the member links to the same destination, thus minimizing the likelihood of network congestion.
Link aggregation is often applied in areas of the enterprise network where high speed connectivity and the potential for congestion is likely to occur. This generally equates to the core network where responsibility for high speed switching resides, and where traffic from all parts of the enterprise network generally congregates before being forwarded to destinations either in other parts of the network, or remote destinations beyond the boundaries of the enterprise network. The example demonstrates how core switches (SWA & SWB) support link aggregation over member links that interconnect the two core switch devices, as a means of ensuring that congestion does not build at a critical point in the network.
Link aggregation supports two modes of implementation, a manual load balancing mode and static LACP mode. In load balancing mode, member interfaces are manually added to a link aggregation group (LAG). All of the interfaces configured with load balancing are set in a forwarding state. The AR2200 can perform load balancing based on destination MAC addresses, source MAC addresses, exclusiveOR of the source and destination MAC addresses, source IP addresses, destination IP addresses, or Exclusive-OR of source and destination IP addresses. The manual load balancing mode does not use the Link Aggregation Control Protocol (LACP), therefore the AR2200 can use this mode if the peer device does not support LACP.
In static LACP mode, devices at two ends of a link negotiate aggregation parameters by exchanging LACP packets. After the negotiation is complete, the two devices determine the active interface and the inactive interface. In this mode, it is necessary to manually create an Eth-Trunk and add members to it. LACP negotiation determines which interfaces are active and which ones are inactive. The static LACP mode is also referred to as M:N mode, where M signifies the active
member links which forward data in a load balancing mode, and N represents those links inactive but providing redundancy. If an active link fails, data forwarding is switched to the backup link with the highest priority, and the status of the backup link changes to active. In static LACP mode, some links may function as backup links, whereas all member interfaces work in a forwarding state in manual load balancing mode, and represents the main difference between the two modes.
As a logical interface for binding multiple physical interfaces and relaying upperlayer data, a trunk interface must ensure that all parameters of the physical interfaces (member interfaces) on both ends of the trunk link be consistent. This includes the number of physical interfaces, the transmission rates and duplex modes of the physical interfaces, and the traffic-control modes of the physical interfaces, for which it should be noted that member interfaces can be layer 2 or layer 3 interfaces. Where the interface speed is not consistent, it is still possible forthe trunk link to operate, however the interfaces operating at a lower rate are likely to experience loss of frames.
In addition, the sequence of the data flow must be unchanged. A data flow can be considered as a group of frames with the same MAC address and IP address. For example, the telnet or FTP connection between two devices can be considered as a data flow. If the trunk interface is not configured, frames that belong to a data flow can still reach their destination in the correct order because data flows are transmitted over a single physical link. When the trunk technology is used, multiple physical links are bound to the same trunk link, and frames are transmitted along these physical links. If the first frame is transmitted over one physical link, and the second frame is transmitted over another physical link, it is possible that the second frame may reach the destination earlier than the first frame To prevent the disorder of frames, a frame forwarding mechanism is used to ensure that frames in the same data flow reach the destination in the correct sequence. This mechanism differentiates data flows based on their MAC addresses or IP addresses. In this manner, frames belonging to the same data flow are transmitted over the same physical link. After the frame forwarding mechanism is used, frames are transmitted based on the following rules:
- Frames with the same source MAC addresses are transmitted over the same physical link.
- Frames with the same destination MAC addresses are transmitted over the same physical link.
- Frames with the same source IP addresses are transmitted over the same physical link.
- Frames with the same destination IP addresses are transmitted over the same physical link.
- Frames with the same source and destination MAC addresses are transmitted over the same physical link.
- Frames with the same source and destination IP addresses are transmitted over the same physical link.
Establishment of Link Aggregation is achieved using the interface Eth-trunk <trunk-id> command. This command creates an Eth-Trunk interface and allows for the Eth-Trunk interface view to be accessed. The trunk-id is a value used to uniquely identify the Eth-trunk, and can be any integer value from 0 through to 63. If the specified Eth-Trunk already exists, it is possible to directly enter the EthTrunk interface view by using the interface Eth-trunk command. An Eth-Trunk can only be deleted if the Eth-Trunk does not contain any member interfaces. When adding an interface to an Eth-Trunk, member interfaces of a layer 2 Eth-Trunk must be layer 2 interfaces, and member interfaces of a layer 3 Eth-Trunk must be layer 3 interfaces. An Eth-Trunk can support a maximum of eight member interfaces. A member interface cannot have any service or static MAC address configured. Interfaces added to an Eth-Trunk should be hybrid interfaces (the default interface type). An Eth-Trunk interface cannot have other Eth-Trunk interfaces as member interfaces. An Ethernet interface can be added to only one Eth-trunk interface.
To add the Ethernet interface to another Eth-trunk, the Ethernet interface must be deleted from the current Eth-Trunk first. Member interfaces of an Eth-trunk must be the same type, for example, a Fast Ethernet interface and a Gigabit Ethernet interface cannot be added to the same Eth-trunk interface. The peer interface directly connected to a member interface of the local Eth-Trunk must also be added to an Eth-Trunk, otherwise the two ends cannot communicate. When member interfaces have different rates, the interfaces with lower rates may become congested and packet loss may occur. After an interface is added to an Eth-Trunk, MAC address learning is performed by the Eth-Trunk rather than the member interfaces.
In order to configure layer 3 Link Aggregation on an Ethernet trunk link, it is necessary to transition the trunk from layer 2 to layer 3 using the undo portswitch command under the Eth-trunk logical interface. Once the undo portswitch command has been performed, an IP address can be assigned to the logical interface and the physical member interfaces that are to be associated with the Ethernet trunk link can be added
Using the display interface eth-trunk <trunk-id> command it is possible to confirm the successful implementation of Link Aggregation between the two peering devices. The command can also be used to collect traffic statistics and locate faults on the interface.
The current state of the Eth-trunk is set to UP, signaling that the interface is operating normally. Where the interface shows as down, this signals that an error has occurred at the physical layer, whereas an administratively down error reflects that the shutdown command has be used on the interface. The specific error in the event of a failure can be discovered by verifying the status of the ports, for which all ports are expected to show an UP status. Load balancing is supported when the weight of all links is considered equal.
A Fast Ethernet interface and a Gigabit Ethernet interface cannot be added to the same Eth-trunk interface, any attempt to establish member links of different types will result in an error specifying that the trunk has added a member of another port-type. It should be noted that the S5700 series switch supports Gigabit Ethernet interfaces only, however this behavior can be applied to other models including the S3700 switch.
Only the LACP mode is capable of supporting backup member links and therefore should be used if backup links are required.