Backplane segmentation, pros and cons

Hi ​@Daniel Martinez 

Good question.
This topic comes up quite often in AHV designs.

In most standard production environments running with 10/25GbE networking, low latency, and properly sized switching, it is generally fine to keep CVM backplane traffic on the same network as the CVM and host data network. Nutanix does not require a dedicated VLAN for CVM-to-CVM traffic such as Stargate or metadata replication, as long as bandwidth and latency requirements are properly met.

Keeping everything on the same network works well when the network is not oversubscribed, east-west traffic is reasonable, and QoS and MTU are correctly configured. In these cases, performance is typically stable and predictable.

Separating the backplane can make sense in very large clusters, environments with heavy east-west replication traffic, or where strict traffic isolation is required by design or policy. It can provide clearer traffic separation and more predictable performance in very busy environments.

However, segmentation also adds design complexity. It introduces additional VLANs and subnets to manage and can make troubleshooting slightly more complex.

In most real-world deployments, especially small to mid-size clusters, a well-designed single network handles CVM backplane traffic without issues.

For more details, refer to the AHV Admin Guide network best practices

Thanks 
Selvamani S

Thanks for sharing your experience.

So in summary segmentation could be interesting to achieve a better performance, in exchange it may provide a bit more network complexity. 

When there is +25Gbps bandwidth and proper designed network all should be working fine without. 

When you have 10Gbps networks and some extra interfaces in the nodes (not being used) then I always segment intra cluster traffic. The added complexity is negligible.

When designing a new cluster I always segment the intra cluster traffic. So when the nodes are ordered there are enough interfaces in it. 

That’s a good strategy...