Distributing load over multiple load balancers

Given a website that could have many hundreds of thousands of concurrent users at peak times how can one manage the laod where load balancer resources can only support 40,000 concurrent connections? (System spec, out of my control)

The only method I can think is to round robbin style redirect all incoming traffic to one of X subdomains (eg. 1.domain.com, 2.domain.com... X.domain.com) where each subdomain points to a different load balancer. Is this kind of redirect and hand off costly from a server perspective? Will it cause browser warnings (eg. page www.domain.com is trying to redirect you to...)? Is there a better method that I'm overlooking? (Execept get a bigger load balancer)

Is there a 'right' way to do this?

A common solution would be to use a Global Server Load Balancer or GEO. In this case user's DNS queries would be distributed to different IP addresses (each mapping to the separate load balancer). Most GLSBs include healthchecks on Load Balancer endpoints (if one fails stop sending users to this IP) and also can be used to route users to the most optimal endpoint where load balancers are spread across the world ie Send Europe Traffic to the Europe end point etc.

This solution would remove the need for the redirect you mentioned which themselves could add unnecessary load to a s

I have few deployment behind me using dynamic routing protocols like OSPF. There was little trigger script written in bash that checked if proxy/app server is working properly and then start/stop bgpd/ospfd (or just add announcement using vtysh). this routing daemon announced one ip address in network (/32 subnet) and then routers split traffic equally if two paths had same cost - it's called ECMP (Equal-cost multipath).