With the growth of Internet popularity and its importance for the business applications staying permanent online becomes a key requirement. Commonly, the broadband access and enterprise networks involve devices with dynamic NAPT(NAT) and stateful firewalls functionality. Keeping open connection behind such an intermediate device is not a trivial task, since the communication participants are not notified for NAPT/Firewall status changes. Until detaching the change at intermediate device and reestablishing the connection, all send packets get lost. To overcome these issues most applications involve deed-peer-detection(updates) mechanisms, which verify the host availability, thus if the connection is active. The fast detection of connection lost is vital for the application and therefore short constant dead-peer-detection intervals are used. The interval value is set based on the general network experience and application requirements. The update interval is a tradeoff between the network resources and fast detection of disconnection. In this paper is suggested a framework for dynamic optimization of the update intervals in respect of fast disconnection detection and low network resources. The key idea is to make the update interval proportional to the probability of connection drop. The main issue is that the time point of disconnection can only be narrowed down to certain update interval, the exact time point can not be determined by the hosts. The proposed solution is based on adaptive fuzzy logic. It is relevant for wide spectrum of applications like VoIP, routing, security protocols etc.