作者: Mark Felegyhazi , Vern Paxson , Christian Kreibich
DOI:
关键词:
摘要: In this paper we explore the potential of leveraging properties inherent to domain registrations and their appearance in DNS zone files predict malicious use domains proactively, using only minimal observation known-bad drive our inference. Our analysis demonstrates that inference procedure derives on average 3.5 15 new from a given domain. 93% these inferred subsequently appear suspect (based third-party assessments), nearly 73% eventually blacklists themselves. For latter, proactively blocking based predictions provides median headstart about 2 days versus reactive blacklist, though gain varies widely for different domains.