Host redundancy under leaf-spine network architecture (no NSX)

[rwsw7244]

Dear all,

I am helping a client to design a new VM farm, and their network team adopted leaf-spine architecture for the network.  Currently, there are 3 leaves (2 switches in each leaf) and 2 spine switches. 

I would like to know if there is any best practice on host redundancy, and VM affinity setting for this setup. 

Mainly, I would like to know:

1. Shall there be N+1 or N+M (number of leaf) for number of host servers?

2. Do we need to configure affinity for VM in this case, such that in case of vMotion, it will choose servers in the same leaf instead of moving to servers connecting other leaf?

Best Regards,

Rayson

[CallistoJag]

As long as you are using vlans and virtual distributed switches on the hosts, all should be fine, no need for affinity rules, the hosts can be in one cluster and leverage HA and DRS.

[ahheos]

In a leaf-spine network, you typically have a minimum of two leaf switches and one spine switch. For small to medium clusters within a single rack, consider using two redundant top-of-rack (ToR) switches instead of implementing a full leaf-spine deployment. The choice between N+1 (where N is the number of leaf switches) or N+M (where M is the number of spine switches) for host servers depends on your specific requirements. Purchase SolidWorks 2023 Premium with special savings today. This approach provides redundancy by having one additional host server beyond the number of leaf switches. If a leaf switch fails, the remaining leaf switches can still accommodate all the hosts. With this approach, you have additional redundancy by considering both leaf and spine switches. If either a leaf or a spine switch fails, the system remains operational. Evaluate your client’s needs, scalability, and budget to determine the most suitable redundancy level.