If for some reason you cannot use LACP, you can also use balance-xor mode to dual-connect host-facing bonds in an MLAG environment. If you do, you must still configure the same clag_id parameter on the MLAG bonds, and it must be the same on both MLAG switches.3 cognitive learning styles otherwise, the MLAG switch pair treats the bonds as if they are single-connected.

In this case, L1 and L2 are also MLAG peer switches, and present a two-port bond from a single logical system to S1 and S2.3 cognitive learning styles S1 and S2 do the same as far as L1 and L2 are concerned. For a switch-to-switch MLAG configuration, each switch pair must have a unique system MAC address.3 cognitive learning styles in the above example, switches L1 and L2 each have the same system MAC address configured. Switch pair S1 and S2 each have the same system MAC address configured; however, it is a different system MAC address than the one used by the switch pair L1 and L2.3 cognitive learning styles LACP and dual-connectedness

For MLAG to operate correctly, the peer switches must know which links are dual-connected or are connected to the same host or switch.3 cognitive learning styles to do this, specify a clag-id for every dual-connected bond on each peer switch; the clag-id must be the same for the corresponding bonds on both peer switches.3 cognitive learning styles typically, link aggregation control protocol (LACP), the IEEE standard protocol for managing bonds, is used for verifying dual-connectedness.3 cognitive learning styles LACP runs on the dual-connected device and on each of the peer switches. On the dual-connected device, the only configuration requirement is to create a bond that is managed by LACP.3 cognitive learning styles

However, if for some reason you cannot use LACP in your environment, you can configure the bonds in balance-xor mode. When using balance-xor mode to dual-connect host-facing bonds in an MLAG environment, you must configure the clag_id parameter on the MLAG bonds, which must be the same on both MLAG switches.3 cognitive learning styles otherwise, the bonds are treated by the MLAG switch pair as if they are single-connected. In short, dual-connectedness is solely determined by matching clag_id and any misconnection will not be detected.3 cognitive learning styles

Each peer switch periodically makes a list of the LACP partner MAC addresses for all of their bonds and sends that list to its peer (using the clagd service; see below).3 cognitive learning styles the LACP partner MAC address is the MAC address of the system at the other end of a bond (hosts H1, H2, and H3 in the figure above). When a switch receives this list from its peer, it compares the list to the LACP partner MAC addresses on its switch.3 cognitive learning styles if any matches are found and the clag-id for those bonds match, then that bond is a dual-connected bond. You can also find the LACP partner MAC address by the running net show bridge macs command or by examining the /sys/class/net//bonding/ad_partner_mac sysfs file for each bond.3 cognitive learning styles configure MLAG

Upon receipt of a valid message from its peer, the switch knows that clagd is alive and executing on that peer. This causes clagd to change the system ID of each bond that is assigned a clag-id from the default value (the MAC address of the bond) to the system ID assigned to both peer switches.3 cognitive learning styles this makes the hosts connected to each switch act as if they are connected to the same system so that they use all ports within their bond. Additionally, clagd determines which bonds are dual-connected and modifies the forwarding and learning behavior to accommodate these dual-connected bonds.3 cognitive learning styles

If the peer does not receive any messages for three update intervals, then that peer switch is assumed to no longer be acting as an MLAG peer.3 cognitive learning styles in this case, the switch reverts all configuration changes so that it operates as a standard non-MLAG switch. This includes removing all statically assigned MAC addresses, clearing the egress forwarding mask, and allowing addresses to move from any port to the peer port.3 cognitive learning styles after a message is again received from the peer, MLAG operation starts again as described earlier. You can configure a custom timeout setting by adding --peertimeout  to clagd-args, like this:

3 cognitive learning styles

In this design, the spine switches route traffic between the server hosts in the layer 2 domains and the core. The servers (host1 thru host4) each have a layer 2 connection up to the spine layer where the default gateway for the host subnets resides.3 cognitive learning styles however, since the spine switches as gateway devices communicate at layer 3, you need to configure a protocol such as VRR (virtual router redundancy) between the spine switch pair to support active/active forwarding.3 cognitive learning styles

Then, to connect the spine switches to the core switches, you need to determine whether the routing is static or dynamic. If it is dynamic, you must choose which protocol — OSPF or BGP — to use.3 cognitive learning styles when enabling a routing protocol in an MLAG environment, it is also necessary to manage the uplinks, because by default MLAG is not aware of layer 3 uplink interfaces.3 cognitive learning styles in the event of a peer link failure, MLAG does not remove static routes or bring down a BGP or OSPF adjacency unless a separate link state daemon such as ifplugd is used.3 cognitive learning styles

The peer link carries very little traffic when compared to the bandwidth consumed by dataplane traffic. In a typical MLAG configuration, most every connection between the two switches in the MLAG pair is dual-connected, so the only traffic going across the peer link is traffic from the clagd process and some LLDP or LACP traffic; the traffic received on the peer link is not forwarded out of the dual-connected bonds.3 cognitive learning styles

In general, you need to determine how much bandwidth is traveling across the single-connected interfaces, and allocate half of that bandwidth to the peer link.3 cognitive learning styles we recommend half of the single-connected bandwidth because, on average, one half of the traffic destined to the single-connected host arrives on the switch directly connected to the single-connected host and the other half arrives on the switch that is not directly connected to the single-connected host.3 cognitive learning styles when this happens, only the traffic that arrives on the switch that is not directly connected to the single-connected host needs to traverse the peer link, which is how you calculate 50% of the traffic.3 cognitive learning styles

In the illustration below, each host has two 10G links, with each 10G link going to each switch in the MLAG pair. Each host has 20G of dual-connected bandwidth, so all three hosts have a total of 60G of dual-connected bandwidth.3 cognitive learning styles we recommend you allocate at least 15G of bandwidth to each peer link bond, which represents half of the single-connected bandwidth.

Scaling this example out to a full rack, when planning for link failures, you need only allocate enough bandwidth to meet your site's strategy for handling failure scenarios.3 cognitive learning styles imagine a full rack with 40 servers and two switches. You might plan for four to six servers to lose connectivity to a single switch and become single connected before you respond to the event.3 cognitive learning styles so expanding upon our previous example, if you have 40 hosts each with 20G of bandwidth dual-connected to the MLAG pair, you might allocate 20G to 30G of bandwidth to the peer link — which accounts for half of the single-connected bandwidth for four to six hosts.3 cognitive learning styles failover redundancy scenarios