I'm currently working  on scale testing octavia loadbalancer with RHOSO on 3 node compact OCP cluster with 37 computes and 9 ceph nodes. I was able to successfully scale upto 7K load balancers without any failure. The scale was performed in batches of 1K load balancers per run during testing. After 7k the load balancers started moving to the error state. Now all the load balancers except around 100 load balancers are active and others moved from active to error state (amphora vm's are still active).