There is a known problem with the Java libraries around sockets that can cause the socket to not clean up correctly.  This is particularly prevalent in high latency environments.  The sockets will time out, but will not be cleaned up.  These sockets will stack up until the integration fails due to its inability to acquire new sockets.  This is also a way to verify the problem.  Check the sockets associated to the RTC server and see how many exist.  Currently there is no workaround short of cycling the server to restart the JVM which will reclaim the sockets, or manually terminating the leaked sockets.  Contact support for a patch to fix the problem if the problem persists and becomes unmanagable.