docs/doc/source/node_management/kubernetes/node_interfaces/link-aggregation-settings.rst

Link Aggregation Settings

supports several link aggregation () operational modes.

If you select link aggregation (also known as aggregated Ethernet) when configuring the management, cluster-host, or networks, you can choose from the following operational modes. For more information, refer to the Linux kernel Ethernet Bonding Driver documentation available online (https://www.kernel.org/doc/Documentation/networking/bonding.txt).

Note

Ensure that the mode on the corresponding switch ports is configured to match your selection.

Table 1. Supported Link Aggregation Operational Modes

Mode	Description	Supported Interface Types
Active-backup (default value)	Provides fault tolerance. Only one standby interface at a time is available. The backup standby interface becomes active only when the active standby interface fails. For platform interfaces (such as, , cluster-host, and management interfaces), the system will select the interface with the lowest address as the primary interface when all standby interfaces are enabled.	Management, , cluster-host, and data interface
Balanced XOR	Provides aggregated bandwidth and fault tolerance. The same standby interface is used for each destination address. This mode uses the default transmit policy, where the target standby interface is determined by calculating the source address XOR'd with the destination address, modulo 2. You can modify the transmit policy using the xmit-hash-policy option. For details, see Table 2 <link-aggregation-settings-xmit-hash-policy>.	, cluster-host, and data interfaces
802.3ad	Provides aggregated bandwidth and fault tolerance. Implements dynamic link aggregation as per the IEEE 802.3ad specification. You can modify the transmit policy using the xmit-hash-policy option. For details, see Table 2 <link-aggregation-settings-xmit-hash-policy>. In order to support booting over an aggregated management interface, the far-end switch ports must be configured in passive mode. This is required because the BIOS on the host does not support and cannot establish a , and therefore can use only one of the aggregated interfaces during boot. If the far-end switch is configured to use active , it can establish a and use either interface, potentially resulting in a communication failure during the boot process.	Management, , cluster-host, and data interface

Table 2. xmit-hash-policy Options

Options	Description	Supported Interface Types
Layer 2 (default value)	Hashes on source and destination addresses.	, internal management, cluster-host, and data interfaces (worker nodes).
Layer 2 + 3	Hashes on source and destination addresses, and on source and destination IP addresses.	, internal management, and cluster-host
Layer 3 + 4	Hashes on source and destination IP addresses, and on source and destination ports.	, internal management, and cluster-host

Table 3. primary_reselect Options

Options	Description	Supported Interface Types
Always (default value)	The primary standby becomes an active standby whenever it comes back up.	, internal management, and cluster-host
Better	The primary standby becomes active standby whenever it comes back up, if the speed and the duplex of the primary standby is better than the speed duplex of the current active standby.	, internal management, and cluster-host
Failure	The primary standby becomes the active standby only if the current active standby fails and the primary standby is up.	, internal management, and cluster-host

LAG Configurations for AIO Duplex Systems

For a duplex-direct system set-up, use a mode with active-backup for the management interface when attaching cables between the active and standby controller nodes. When both interfaces are enabled, the system automatically selects the primary interface within the with the lowest address on the active controller to connect to the primary interface within the with the lowest address on the standby controller.

The controllers act independently of each other when selecting the primary interface. Therefore, it is critical that the inter-node cabling is completed to ensure that both nodes select a primary interface that is attached to the primary interface of the opposite node. The inter-node management cabling attachments must be from the lowest address to the lowest address for the first cable, and the next lowest address to the next lowest address for the second cable. Failure to follow these cabling requirements will result in a loss of communication between the two nodes.

In addition to the special cabling requirements, the node BIOS settings may need to be configured to ensure that the node attempts to network boot from the lowest address interface within the . This may be required only on systems that enable all hardware interfaces during network booting rather than only enabling the interface that is currently selected for booting.

Configure the cables associated with the management so that the primary interface within the with the lowest address on the active controller connects to the primary interface within the with the lowest address on standby controller.