Traffic Distribution Algorithm for N-Series LAGs

  • 1.  Traffic Distribution Algorithm for N-Series LAGs

    Posted 11-25-2013 12:49
    Article ID: 12248

    Matrix N-Series DFE

    Few of the available LAG paths are being used.
    LAG is not load balancing.

    On the DFE, 802.3ad Link Aggregation permits the bundling of an effectively unlimited number of physical ports (within one system) into a single logical port, providing for increased bandwidth and redundancy. The load is distributed across the links in a manner compliant with the IEEE 802.3 specification. All DFE models support up to 48 such trunk groups per system, with the individual links distributed anywhere within the system.

    When each packet traverses a LAG port (ex: lag.0.1), the actual underlying port selection is made based upon the following:
    • Broadcast, non-IGMP multicast, and unlearned unicast traffic uses the lowest port number for slow-path (CPU intensive) forwarding.
    • This leaves IGMP multicast and learned unicast traffic, all of which is programmed into hardware for fast-path forwarding, and is mapped depending upon the LAG algorithm selected by means of the 'set lacp outportAlgorithm <option>' command, where <option> is one of: * dip-sip Use Source IP / Dest IP for outport determination [default] * da-sa Use Dest MAC / Source MAC for outport determination * round-robin Round-robin the flows for outport determination[/code]
    When using the dip-sip or da-sa algorithm, flows are first mapped to a number within the range of 0-255, then this is reduced to a number matching a link instance which in turn maps to an underlying physical port.

    When using the round-robin algorithm, flows are mapped to each successive underlying port in ascending port sequence. A flow is defined as a unique source and destination IP address (L3), and possibly a unique UDP,TCP pair (L4). By default, flows are created up to L3. Awareness of flows through L4 is attained by configuring at least one feature which examines L4 data. The use of NetFlow, LSNAT, NAT, L4 ACLs, and/or L4 Policies would do this.

    Hint: Changing the algorithm will not typically affect existing flows which are already mapped to an "outport". The workaround is to enable flow regeneration ('set lacp flowRegeneration enable'), and then unlink/relink any physical LAG ports to which you would like some existing flows to be mapped. A remap will only then occur if the port(s) in question would normally be the target of equivalent new flows, but a remap will not even otherwise be considered for existing flows.

    See also: 5203 and 14165.