Aggregated Ethernet Interfaces

You can group or bundle multiple Ethernet interfaces together to form a single link layer interface known as the aggregated Ethernet interface (aex) or a link aggregation group (LAG). The IEEE 802.3ad standard defines link aggregation of Ethernet interfaces and provides a method by which you can group or bundle multiple Ethernet interfaces. Bundling multiple interfaces together enables you to increase the supported bandwidth. The device treats the aggregated Ethernet interface or LAG as a single link instead of a combination of multiple links.

Consider the following guidelines as you configure an aggregated Ethernet interface.

• For Junos OS Evolved, if you add a new member interface to the aggregated Ethernet bundle, a link flap event is generated. The physical interface is deleted as a regular interface and then added back as a member. During this time, the details of the physical interface are lost.

• You must not configure aggregated Ethernet for subscriber management by using the ether-options statement. If you do so, subscriber management does not work properly—there are issues with subscriber accounting and statistics. Use the gigether-options statement to configure aggregated Ethernet interfaces on the member link interfaces.

• You cannot configure simple filters on member link interfaces in an aggregated Ethernet bundle.

• You cannot configure any IQ-specific capabilities such as MAC accounting, VLAN rewrites, or VLAN queuing on member link interfaces in an aggregated Ethernet bundle.

On Juniper Networks devices, you can configure the member links of an aggregated Ethernet bundle to operate at different link speeds (also known as rates). The configured aggregated Ethernet bundle is known as a mixed-rate aggregated Ethernet bundle. When you configure the member links of an aggregated Ethernet bundle in LAN mode as well as WAN mode for 10-Gigabit Ethernet interfaces, the configuration is known as mixed-mode configuration.

Table 5 lists the platforms and corresponding MPCs that support mixed-rate aggregated Ethernet bundles on MX Series routers.

For information about devices that support Mixed Aggregated Ethernet Bundles in Junos OS and Junos OS Evolved see Feature Explorer.

Table 6 lists the platforms and corresponding hardware components that support mixed aggregated Ethernet bundles.

Consider the following guidelines as you configure a mixed-rate aggregated Ethernet bundle:

• You can configure a maximum of 64 member links to form a mixed aggregated Ethernet bundle.

• When you mix a 10-Gigabit Ethernet interface in LAN mode and a 10-Gigabit Ethernet interface in WAN mode in the same aggregated bundle on MX Series, it is not considered a mixed-rate aggregate. To mix the interfaces having the same speed but different framing options, you need not use the mixed statement at the [edit interfaces interface-name aggregated-ether-options link-speed] hierarchy level.

• Mixed-rate aggregated Ethernet links can interoperate with non-Juniper Networks aggregated Ethernet member links provided that mixed-rate aggregated Ethernet load balancing is configured at egress.

• After you configure a mixed-rate aggregated Ethernet link on a 100-Gigabit Ethernet PIC with CFP, changing aggregated Ethernet link protection or LACP link protection configurations results in aggregated Ethernet link flapping. Also, changing the configuration of a mixed aggregated Ethernet link can result in aggregated Ethernet link flapping.

• Packets are dropped when the total throughput of the hash flow exiting a member link (or the throughput of multiple hash flows exiting a single member link) exceeds the link speed of the member link. This can happen when the egress member link changes because of a link failure and the hash flow switches to a member link of speed that is less than the total throughput of the hash flow.

• Mixed-rate aggregated Ethernet links do not support rate-based CoS components such as scheduler, shaper, and policer. However, the default CoS settings are supported on the mixed-rate aggregated Ethernet links.

• Load balancing of the egress traffic across the member links of a mixed-rate aggregated Ethernet link is proportional to the rates of the member links. Egress multicast load balancing is not supported on mixed aggregated Ethernet interfaces.

• Mixed-rate aggregated Ethernet interface do not support aggregated Ethernet link protection, link protection on a 1:1 model, and LACP link protection.

By default, LACP is not configured on aggregated Ethernet interfaces. Ethernet links do not exchange information about the state of the link. When you configure LACP, the transmitting link (also known as actor) initiates transmission of LACP packets to the receiving link (also known as partner). The actor is the local interface in an LACP exchange. The partner is the remote interface in an LACP exchange.

When you configure LACP, you must select one of the following transmission modes for each end of the LAG:

• Active-To initiate transmission of LACP packets and response to LACP packets, you must configure LACP in active mode. If either the actor or partner is active, they exchange LACP packets.

• Passive-There is no exchange of LACP packets. This is the default transmission mode.

Consider the following guidelines when you configure LACP:

• When you configure LACP on multiple different physical interfaces, only features that are supported across all of the linked devices are supported in the resulting link aggregation group (LAG) bundle. For example, different PICs can support a different number of forwarding classes. If you use link aggregation to link together the ports of a PIC that supports up to 16 forwarding classes with a PIC that supports up to 8 forwarding classes, the resulting LAG bundle supports up to 8 forwarding classes. Similarly, linking together a PIC that supports weighted random early detection (WRED) with a PIC that does not support it results in a LAG bundle that does not support WRED.

• If you configure the LACP system identifier (by using the system-id systemid statement) to be all zeros (00:00:00:00:00:00), the commit operation throws an error.

• When you enable a device to process packets received on a member link irrespective of the LACP state if the state of the aggregated Ethernet bundle is up (by using the accept-data statement), then the device does not process the packets as defined in the IEEE 802.3ax standard. According to this standard, the packets should be dropped, but they are processed instead because you configured the accept-data statement.

Targeted distribution provides a mechanism to direct traffic through specified links of an aggregated Ethernet bundle. You can also use targeted distribution to assign roles to member links to handle link failure scenarios. Targeted distribution ensures accurate policy enforcement that is not distributed for a given logical interface. Targeted distribution is applicable to both Layer 2 and Layer 3 interfaces, irrespective of the family configured for the logical interface. The outbound traffic of a Layer 3 host is distributed among all the member links of an aggregated Ethernet bundle. Targeted distribution is implemented only for the transit traffic.

You can form distribution lists consisting of member links of the aggregated Ethernet interfaces and you can assign roles to these lists, as follows:

• Primary distribution list: You can configure the member links that will be part of the primary distribution list. Traffic is load-balanced among all the member links in the primary list. If all links within the primary list are up, traffic is forwarded on those links. If some of the links within a primary list fail, the remaining links carry traffic.

• Backup distribution list: You can configure the member links that will be part of the backup distribution list. If all links within the primary list go down, only then the links in the backup list start carrying traffic. If some of links within the backup list fail, the remaining links in the backup list carry traffic.

• Standby distribution list: All remaining links are added to the defined standby list. If all the links within the primary list and the backup list go down, only then the links in the standby list start carrying traffic. When the links in the primary distribution list come back online, they resume carrying traffic.

This example shows how to configure primary and backup targeted distribution lists for aggregated Ethernet member links. Member links are assigned membership to the distribution lists. Logical interfaces of the aggregated Ethernet bundle are then assigned membership to the primary list and the backup list.

By default, dynamic learning of source and destination MAC addresses on aggregated Ethernet interfaces is disabled. When you enable this feature, you can configure source and destination MAC address-based accounting on the routed interfaces on MX Series routers with DPCs and MPCs. Also, when you enable dynamic learning of MAC addresses, the MAC-filter settings for each member link of the aggregated Ethernet bundle is updated. The limit on the maximum number of MAC addresses that can be learned from an interface does not apply to this dynamic learning of MAC addresses functionality.

Destination MAC-based accounting is supported only for MAC addresses dynamically learned at the ingress interface, including each individual child or member link of the aggregated Ethernet bundle. MPCs do not support destination MAC address learning. Dynamic learning of MAC addresses can be supported on only the aggregated Ethernet interface or on selective individual member links. MAC learning support on the bundle depends on the capability of individual member links. If a link in the bundle does not contain the capability to support MAC learning or accounting, it is disabled on the aggregated Ethernet bundle.

The MAC data for the aggregated bundle is displayed after collecting data from individual child links. On DPCs, these packets are accounted in the egress direction (Output Packet/Byte count), whereas on MPCs, these packets are not accounted because DMAC learning is not supported. This difference in behavior also occurs between child links on DPCs and MPCs. Because this feature to enable dynamic learning is related to collecting MAC database statistics from child links based on the command issued from the CLI, there is an impact on the time it takes to display the data on the console based on the size of the MAC database and the number of child links spread across different FPCs.

When you configure enhanced LAG, child next hops are not created for member links and, as a result, a higher number of next hops can be supported. To configure enhanced LAG, you must configure the device’s network services mode as enhanced-ip. This feature is not supported if the device’s network services mode is set to operate in the enhanced-ethernet mode. This feature is enabled by default if the network services mode on the device is configured as enhanced-mode.