Just wondering whether people think it'd be a good idea to have a kicking mechanism at the transport level (i.e. no originating message in some time indicates the client is idle and should be culled, as opposed to no network / packet activity)? To differentiate: messages are meaningful, packets may or may not be a complete message.

I would suggest you do the boots at the command process level. Set a time for the last valid command, then check the time perododically.

theTroll

It might be useful if the transport delivers some notification when it decides that the other end is un-cooperative. However, the decision of whether to kick or not should probably be left to the application. The notification might state whether it's been getting no data, incomplete data, corrupt data, etc.

I think I'll go with hplus's suggestion, and allow the app to set an idle timeout after which the transport layer will generate a 'has gone idle' event.