Previouse behavior:
In the NSX version 4.1 the pathing from south to north traffic always takes a single TAP interface on the edge when it goes to the physical side. It is because it hasonly one backplane switch that runs between the hypervisor services runningon a NSX DR and the SR component running in the hypervisor all the way to theedge VM.
NSX 4.2 changes:
From NSX 4.2 that concempt was changed and it called TEP, which provides more than one path from the each host tothe edge VM. Gives us a full symmetric bandwidth from south tonorth and north to south.
North to south was always multi pathing but from now the new feature means that actually we can utilize both of ourPNIC and the edge performance is heavily dependent upon how many PNICs we have.
Each PNIC has its own resource queues in the ESXihypervisor that creates the threading. Threading gives us a capability to queuethe packet simultaneously and run the scheduler right. So it has a directimplications on as many queues and as many PNICs we can use for the givenpath.
This is available on SDDCmanager as well as on a standalone NSX deployment and represented as API call as we don't have a GUI yet. GUI is typically run behind so VMware always do API calls first and then introduce things on the GUI the later.
- Centralized routing leverages a single TEP of the edge where the gateway is instantiated.
- This could lead to bandwidth constraints for demanding customers.
- The new feature bundles the TEPs of an edge into a TEP group.
- Traffic from/to the TEP group is L4 hashed across the members of the group, thus providing high bandwidth and availability thanks to fine grained load sharing.
Enable the feature with an API call:
GET: https: //<NSX manager>/policy/api/v1/infra/connectivity-global-config
{ ...
"tep_group_config": {
"enable_tep_grouping_on_edge": false → true
}, ...
}
PUT: https://<NSX manager>/policy/api/v1/infra/connectivity-global-config
