부산대학교 도서관

We present a new technique for detecting the possible presence of certain Denial-of-Service attacks in network traffic. The effectiveness of the technique is demonstrated in experiments against 23 attacks in three different traffic backgrounds. Even though some attacks persist for only 2-4 seconds, results show detection rates up to 84%. Results also show that (for these data) the technique can be tuned to eliminate false alarms. These results are especially favorable given the technique's objective of detecting new (previously unseen) attacks without a template of the background traffic.