This will be at the SignalR level. A few design challenges:

• This is unneeded for long polling but SignalR doesn't know what the underlying transport is (it could sniff connection metadata)
• Do we treat it like a regular invocation message with a specific method name? Payload might be bigger than it would need to not conflict with existing Hub method names.
• Is it in both directions? In SignalR 2.0 we did server to client keep alives to stop proxies from killing the connection. We at least need to do that work but do we need client to server keep alives?
• How often is it sent?