I have many linux servers (cluster) to run same application.Application use C/S structure,client connect to server using TCP long connection(server is basing apache mina socket framework). my question is : When one server shutdown, how other servers can keep the socket connection established between the failure server and the clients?
so the server-down failure is transparent to client and the client need not reconnect to server?

Thanks
L.J.W

sounds like you need a load balancer in the middle there, and some sort of session mirroring on the back? How do you currently control which clients hit which servers? Generally taking over the live TCP connection is not going to happen, but with a load balancer between them, it can handle the broken connection on the server side and re-establish unknown to the real client. Totally depends if the internals of the application can actually handle this. With a connection lasting so long, it sounds like there must be a lot of connection specific data. Is it at all possible to just start sending existing data to a new server and it have a clue what's going on?

Thanks for reply,I want to use LVS to implement cluster,but It seems it can not solve the long TCP connection keeping issue.any solution can be failover to client and application transparent?