Information on the ECMP behavior.
VMware NSX-T Data Center 3.x
VMware NSX 4.x
ECMP hashing is based on a 5-tuple algorithm that uses source IP address, destination IP address, source port, destination port, and IP protocol.
This method allows a better distribution of the traffic across all the available paths. Due to the randomness of the hashing algorithm this behavior is caused.
This is expected behaviour.
Workaround:
1. One option is to connect Tier-1 to Tier-0 on the same edge cluster as Tier-0. This way traffic can be marked differently and balanced across all uplinks.
2. Ensure the number of edge nodes in Tier-0 cluster (used by Tier-1's ECMP) is not same as the number of Tier-0 uplinks (used by Tier-0's ECMP). To address this, create an additional uplink for Tier-0 to diversify the usage of Tier-0 uplinks. This ensures a more balanced distribution of traffic across the network, improving overall network efficiency.
Issue with the ECMP could be seen in some of the following cases:
1. If Tier0 and tier1 are on a different edge cluster.
2. Each tier0 has two uplinks, and 2 edges in the edge cluster and the ECMP to north external router is via uplink1 and uplink2 for edge node 1 and via uplink 3, uplink 4 for edge node 2 - which makes 2-way ECMP per edge node. At the same time, the Tier-1 is also having a 2-way ECMP to both edge nodes.
3. Tier-1 is connected to Tier-0 edge cluster which only has two edge nodes, so the ECMP from Tier-1 to Tier-0 is via Tier-0 Edge A and Tier-0 Edge B, which is 2-way ECMP.
To correct the above configuration, make either of the following topology changes below: