Tuning Tomcat for a High Throughput, Fail Fast System

> If the number of concurrent requests is near the estimated active threads (8 in our example) then return an http status code of 503.

Once that happens, what are the options? Does a LB forward the request to another app server? Return a 503 to the end-user?

I've had clients that have been unwilling to see the point of view that throttling the workload is, in the end, the right way to go, rather than continuing on and probably eventually hitting total resource exhaustion.