Latency-500K/8: CPU Cost and Queueing at 500K req/s on Four Cores

Half a million requests per second, offered at a fixed rate to a server that has four cores and their SMT siblings to serve it with. Latency-1M gives every entry thirty-two cores at a load only the fastest can feel; this gives it eight logical CPUs at a load that keeps them busy, so what separates them is the cost of a request when the machine is actually busy, and the queue each one builds to keep up. Reference-only while the first entries are measured.