Abstract: It is difficult to detect the whole IPv6 deployment because of the massive address space. Based on the IPv6 unicast address structure, hitlist mechanism is adopted to solve this problem. First, the IPv6 addresses from public ipv6 address data sources are collected, cleaned and stored in the database according to structure of the original IPv6 address set. After that, the dependency between the stable and unstable segments of the address set is studied. An IPv6 address survivability prediction model based on the generation adversarial network are proposed. By considering the discrete nature of IPv6 addresses structure, deterministic binomial neurons and random binomial neurons are designed, and Sigmoid is used to adjust the pass-through estimator to solve the problem of high computational complexity of the back propagation of binomial neurons. The IPv6 address set is trained in each specific AS domain, and the obtained generator is used as an IPv6 address survivability prediction model to generate an IPv6 address hitlist. Experimental analysis results and internet probe results show that the proposed model can effectively generate new IPv6 addresses, and the probability of getting an ICMPv6 response in actual detection is higher than that of sequential or random scans. The survival rates have great difference from 0 to 61%.

Key words: IPv6 address, generative adversarial network, hitlist