Actor Model Anomaly Detection Using Kernel Principal Component Analysis

With the increasing complexity of Internet applications, traditional software architectures have been unable to support the pressure of system access brought about by user growth. Distributed systems have gradually become the mainstream architecture, and messaging has become a widely adopted model. Akka is a distributed framework based on the Actor message communication model. At present, the fault and anomaly detection for the Actor system is mainly to capture the anomaly in the code writing, it is difficult to decouple from the program, so an algorithm using kernel principal component analysis algorithm based on message monitoring is proposed to detect anomaly on Actor system. In this paper, we obtain the message of Actor system by using AspectJ's slicing of the byte code injection of Java code, and we can use Kernel Principal Component Analysis algorithm to perform data dimension reduction and feature extraction through nonlinear mapping. Then the k-means algorithm was used for cluster analysis. The LOF (local outlier points factor) algorithm was used to compare the density of each point p and its neighborhood points to determine abnormal points. Finally, we took the spider program based on the Actor model as a case to collect data and do the experiment, which verified the validity and rationality of the method.