Help us improve your experience.

Let us know what you think.

Do you have time for a two-minute survey?

Navigation
Guide That Contains This Content
[+] Expand All
[-] Collapse All

    QoS for 802.3ad Link Aggregation Interfaces Overview

    You can configure QoS for 802.3ad link aggregation interfaces. To ensure that QoS is applied properly to the interface column, you configure the QoS profile using either a hashed loadbalancing scheme or a subscriber loadbalancing scheme.

    Types of Load Balancing

    For hashed load balancing, you configure the scheduler hierarchy with Ethernet queues, and the system replicates them on each link within the link aggregation group (LAG). The system demultiplexes each packet to one of the active links in the LAG using a random hash generated by fields in the packet header. For example, when an IP packet is routed to a LAG, the hash algorithm is based on the IP Source Address and Destination Address in the IP header.

    For subscriber load balancing, you configure the scheduler hierarchy with IP, VLAN, and S-VLAN queues and the system allocates them to individual ports in the LAG. The system demultiplexes each packet to an active link based on the subinterface underlying the egress interface. For example, when an IP packet is routed to an IP interface over a LAG, the system binds the underlying VLAN, PPPoE, or MPLS subinterface to one of the active links in the LAG. The packet is transmitted over the interface.

    Most network operators configure QoS over 802.3ad LAGs using subscriber load balancing to take advantage of subscriber class-based queuing (SCBQ) features. However, configuring hashed load balancing is useful for achieving fine-grained distribution of multicast VLAN traffic or for any high bandwidth VLAN that does not require shared shaping.

    To ensure that QoS is symmetrically applied to all the links, the router periodically rebalances the traffic on the LAG. You can control the loadbalancing parameters.

    If you configure hashed load balancing to specify the scheduler hierarchy with Ethernet queues and enable the system to replicate them on each link within the LAG, traffic that is transmitted through the LAG bundle might not be evenly distributed across all the member interfaces in the LAG. For example, if a LAG bundle contains two Gigabit Ethernet member interfaces, the traffic that is sent through the LAG bundle might not be equally balanced between the two interfaces. This method of load balancing is expected. The distribution depends on the capability of the router to distribute the traffic with an IP source address/destination address hashing algorithm. Depending on the random nature of the traffic, the traffic is distributed. The hashing algorithm validates the second and fourth octets of the source and destination addresses. Depending on the traffic patterns, the end result might be unevenly balanced use of the interfaces involved in the LAG.

    The algorithm used by the router for forwarding over a LAG bundle might cause a distribution of the traffic that is less then equal. The algorithm operates by creating eight bins, numbered 0-7. There are always eight bins, regardless of the number of interfaces in the bundle. The eight bins are distributed across the links based on an L2 channel algorithm. Each link is allocated to one of the eight bins. When one of the links in the LAG bundle fails, the traffic at that point is not equally distributed across the member links. Whenever an odd number of links are present in a LAG interface, such an imbalanced distribution occurs.

    Munged QoS Profiles and Load Balancing

    To determine whether to use hashed load balancing or subscriber load balancing, the system munges a QoS profile for a subscriber.

    In typical Ethernet configurations, the munged QoS profile for a given subscriber interface comprises the accumulated rules of the QoS profiles attached below the subscriber interface in the interface column. Rules in higher-attached QoS profiles override or eclipse rules in lower-attached QoS profiles. For example, rules from specific interface attachments such as a VLAN override those from attachments at S-VLANs or ports.

    When applying QoS to LAGs, the system uses a modified algorithm to munge QoS profile attachments. The system automatically builds the munged QoS profile using the rules in the QoS profile attached at the LAG interface.

    For example, the munged Qos profile for VLAN 0,0 consists of the munge of:

    • Attachment 1—QoS profile attached to the VLAN
    • Attachment 2—QoS profile attached to the S-VLAN
    • Attachment 3—QoS profile attached to the LAG

    If there is no QoS profile attached to the LAG, the system locates the lag-default QoS profile indicated in the qos-port-type-profile command.

    If the resulting QoS profile specifies only Ethernet queues, the system uses the hash algorithm to balance the links. If the resulting QoS profile specifies any VLAN, IP, or L2TP-Session queues, then the system uses subscriber load balancing.

    802.3ad Link Aggregation and QoS Parameters

    You can create parameter instances for IEEE 803.ad LAG interfaces. A parameter instance for LAG can control an Ethernet port or a node, but you cannot create parameter instances for the Ethernet interfaces within the LAG.

    For example, a LAG instance can specify a shaping rate of 100 Mbps on an Ethernet port or a group node. The system shapes all Ethernet ports or group nodes to the same rate within the LAG. Using load balancing, the system strives to balance the traffic each link equally.

    QoS and Ethernet Link Redundancy

    You can configure Ethernet link redundancy for LAG interfaces. When you configure QoS for those links, be sure to consider the following behaviors.

    Active Link Failure and QoS

    When an active link fails, traffic that is hashed-load balanced is redirected onto the remaining active links in the LAG. Traffic that is hashed-load balanced might be lost on the disabled link, but from the moment of switchover, traffic arriving from the fabric on the egress line module is directed towards one of the remaining hashed load-balanced queues.

    Subscriber loadbalanced traffic takes more time to reestablish on active links because of the amount of computation (approximately 3 ms per subscriber). During this time period, traffic directed to the disabled link might be lost.

    Administratively Disabling a Link and QoS

    When a link is administratively disabled, the system immediately redirects traffic from the link to other links in the LAG.

    Adding a New Link to the LAG and QoS

    When you add a new link to the LAG, the system immediately sends traffic that is hashed-load balanced to the link. Traffic that is subscriber-load balanced moves to the new link as new subscribers log in. The system automatically rebalances traffic to the new link based on the load rebalance configuration for the LAG.

    Published: 2014-08-11