We currently have a windows-service remoting server that is in single-call
mode, and a web-based client application that connects via a TCP channel.
The server works fine for local deployments and remote deployments, and now
we have introduced Windows 2003 network load balancing into the equation, so
that requests to the server are distributed across six machines, each running
a server instance.

Although NLB is doing its job fine, we would now like to implement some form
of reliable failover so that if a remoting server fails and stops listening
on the load-balanced TCP port it is allocated to, the client / NLB cluster
will stop sending requests to it. Currently the only way to do this seems to
involve drainstopping or stopping the affected server node via cluster
administrator, and in the meantime most client requests are dying. Is there
any way to make NLB service-aware, or get it to stop sending requests to a
server that isn't listening on the designated TCP port?

We've added the new 2.0 tcp client channel configuration properties
"socketCacheTimeout" and "socketCachePolicy" to ensure that the client-side
TCP conncetion cache isn't being used, however they are poorly documented.
Does anyone have experience of these?

Thanks in advance