Each appliance registers itself with the routers specified in its service class definitions. The first appliance to register itself, becomes the designated cache, and works with the routers to apportion traffic between itself and the other caches (called subordinate caches). Because your appliances use identical WCCP algorithms, it does not matter which one becomes the designated cache.
As additional appliances come online, they are added to the WCCP cluster, and the traffic is reapportioned among the active appliances. This happens at ten-second increments. After a cold start of the routers or appliances, all of the appliances might come online within the same ten-second window, or they might arrive over multiple ten-second windows, causing traffic to be reapportioned multiple times before it stabilizes. In the latter case, the appliances that come online first maycan become overloaded until additional appliances come online.
An accelerated connection fails when allocated to a different appliance, making reallocation disruptive. This is not true of non-accelerated connections, which generally experience a delay of thirty seconds or more, and then continue. The least-disruptive mapping option minimizes the amount of reallocation when an appliance fails.
If an appliance fails or otherwise goes offline, its absence is noted, and the designated cache reapportions its traffic to the remaining appliances. If the designated cache itself goes offline, the role of designated cache is also reapportioned. It takes about thirty seconds for the cluster to react to the loss of a cache.