Web Caching and Replication

16.3. Server Selection with Multiple Metrics

Many server-selection algorithms rely on a combination of several metrics. One should not confuse these algorithms with those using aggregate metrics: the latter algorithms use a single metric, albeit one that reflects several aspects of system performance. A typical example using multiple metrics is the approach to server selection in which the algorithm selects the nearest server to the client (a proximity metric) provided overloaded servers are not selected (a load metric). We refer to algorithms that follow this general approach as proximity-load-threshold algorithms.

The most natural proximity-load-threshold algorithm is one that considers the servers in order of increasing distance from the client ...