Changing bonding mode in AHV from active-passive to balance-tcp | Nutanix Community
Skip to main content

Let’s say you want  high availability at the NIC level  and failover capacity to the CVM and user VMs.

We recommend using  balance-slb to take advantage of all adapters.

Balance-tcp is a load balancing method that increases host and VM bandwidth utilization beyond a single 10 Gb adapter by balancing each VM NIC TCP session on a different adapter. Also used when network switches require LACP negotiation."

 

Advantages of Bond Mode for balance-slb

  • Balance-slb bond mode in OVS (Open vSwitch) takes advantage of all the links.

  • Rebalance VM traffic from the interface that is used extensively to less used interfaces. When the configurable bond-rebalance interval expires, OVS uses the measured load for each interface and the load for each source MAC hash to spread traffic evenly among links in the bond. 

  • Traffic from some source MAC hashes may move to a less active link to more evenly balanced bond member utilization. 

 
Here is an article that describes the impact of the change and the steps on how to make the network changes: AHV | Configuring Load Balancing active-backup and balance-slb modes

 

Notes: Before making any network changes, please be sure the cluster can tolerate a single node failure. This can be done by checking from the Prism's Home page for Resiliency status. It must be showing a Green OK. Otherwise, please do not make any changes until the resilience issue is resolved. It is highly recommended to do it during a maintenance window after hours.

 

The Nutanix Solutions team has just released a new best practices guide specifically designed to answer advanced questions about AHV networking on the Enterprise Cloud Platform.


The networking described in the AHV Best Practices Guide covers a wide range of the scenarios that Nutanix administrators encounter every day. However, when the defaults don't match customer requirements, you’ll want to use this advanced networking guide.

Hello,

We are looking at expanding our network capabilities beyond A-P/A-B. 

 

In our testing the balance-slb had the adverse effect of creating table-overflows on the switches.  It seemed the CISCO switches were able to self-flush these tables but the Junipers were not (could be down to a ‘policing policy’).

 

Have you experienced this?

 

In the meantime we will review the current documentation.

 

Thank you...