Early Abnormal Overload Detection and the Solution on Content Delivery Network

Abstract

From articles of H. Yu Chen about early detection of network attacks [1], the authors applied his approach to Early Abnormal Overload Detection (EAOD) on Content Delivery Network (CDN) and suggested solutions for the problem, to limit abnormal overload to be occurred on a large network, ensuring users always being accessed to desired resources. Early overload detection mechanism are based on three levels: at each router, in each autonomous system domain (AS domain) and on inter-autonomous domains (inter-AS domains). At each router, when abnormal load exceeds a threshold, it will notify to a server that contains the Change Aggregation Tree (CAT) in the autonomous domain. Each node of the tree is an overloaded router above. On inter-AS domains, the CAT servers exchange information with each other to create global CAT. Based on the height and shape (dense) of the global CAT tree, the overload can be determined on which destination network and which user network caused this overload. Next, the administrator decided to move the content (as a service) which causes overload, to a user network. By this way, it prevents overload on intermediate and destination networks. This approach asks the cooperation among network providers.