TCP/IP is a sensitive traffic. TCP/IP socket connections between 2 processes that run on different machines will be dropped immediately if a request made by the sending process is not acknowledged within a very limited timeframe by the receiving process.
TCP is by definition not a "sensitive" protocol. It's entire purpose is to deal with packet loss, delays, buffer handling etc. to ensure a reliable connection stream of data for the upper application layer protocols. Modern applications and OSes typically have TCP-timeouts between multiple seconds and sometimes minutes, way more time than typically required to stun/unstun a VM.
A snapshot stun/unstun cycle is completely transparent to the network stack, I have never seen "dropped" TCP connections and can't imagine why it should happen.
You may lose a ping or 2, but this kind of delay is nothing special and entirely the reason your application is relying on TCP to deal with such cases. Basiscally it will cause the same kind of network "impact" as a vMotion.
Just test it by connecting via RDP/SSH to your server, take a snapshot, you most likely won't notice anything and your connection will not be dropped - because the underlying TCP session is still fine.
--
http://alpacapowered.wordpress.com
.png "linotelera"){kind=avatar}
