Current Octavia Amphorae scale has been testing with up to 1000 amphorae instances. 

There is a need to benchmark the Octavia scale with RHOSO with higher number of octavia LBs. 

Goal Benchmark Octavia LB scale 

Target

Total LBs scale across the cluster - 20k

Suggested compute node scale - 500 nodes, 27k VMs

While compute node scale and VM scale for RHOSO has been done separately, it is important to understand and verify if the desired LB scale can be achieved with the higher scale for compute nodes and VMs per compute nodes to mimic the customer deployments. 

Suggested amphora VM footprint -95% of LBs: 1 vCPU, 2GB RAM, 5% LBs : 2 vCPUs, 4 GB RAM

• Virtual compute nodes may be used
• Focus here is to validate the control plane and management plane for Octavia and not the data plane
• Functional data plane validation with small amount of traffic crossing the LB must be done
• Since the focus is on control plane scale, validation with kernel data path only is sufficient
• Dynamic control plane testing at target LB scale must be performed (Create, update, delete) to understand the time taken for life-cycle events at large scale

LBs scale may be done incrementally or using binary search. If no issues are observed for the target scale, few iterations of testing with higher scale should be done to find the max LB scale