Performance-based routing

In many-server systems with heterogeneous servers, the Fastest-Server-First (FSF) policy is known for its excellent performance. However, when service rates are unknown and/or time-varying, implementing FSF routing is not straightforward. We analyze an algorithm that approximates FSF routing: servers are ranked in a dynamic list, where the shorter the actual service times that a server exhibits—the closer the server is to the head of the list; a customer is then routed to the lowest-index (highest-in-the-list) idle server.