wan_optimizer,
Multiple keepalive messages are sent by the Interceptor to each Steelhead in the cluster at a frequency of one per second. It takes three consecutive non-responses from any Steelhead for the Interceptor to declare that Steelhead dead. So the entire process takes about three seconds before the Interceptor re-directs traffic from a failed Steelhead.
Note that keepalives are sent at multiple levels, including one from the kernel, as well as an application-level keepalive. The kernel-level keepalive detects a dead box, while the application-level keepalive detects a situation where a Steelhead enters a zombie state, where the box is still alive but the sport optimization process is hung.
Note that this makes the Interceptor far more responsive than WCCP in its default configuration, where keepalives are sent every 10 seconds, and three consecutive non-responses are needed for WCCP to declare a cache engine dead (so the entire process takes more than 30 seconds).
Josh
