On attack correlation and the benefits of sharing IDS data

Abstract

This thesis presents the first wide-scale study of correlated attacks, i.e., attacks mounted by the same source IP against different networks. Using a large dataset from 1700 intrusion detection systems (IDSs), this thesis shows that correlated attacks are prevalent in the current Internet; 20% of all offending sources mount correlated attacks and they account for more than 40% of all the IDS alerts in our logs. Correlated attacks appear at different networks within a few minutes of each other, indicating the difficulty of warding off these attacks by occasional offline exchange of lists of malicious IP addresses. Furthermore, correlated attacks are highly targeted. The 1700 DSs can be divided into small groups with 4-6 members that do not change with time; IDSs in the same group experience a large number of correlated attacks, while IDSs in different groups see almost no correlated attacks These results have important implications on collaborative intrusion detection of common attackers. They show that collaborating IDSs need to exchange alert information in realtime. Further, exchanging alerts among the few fixed IDSs in the same correlation group achieves almost the same benefits as collaborating with all IDSs, while dramatically reducing the overhead.