SOM-based Aging Detection for Virtual Machine Monitor

A virtual monitor machine (VMM) inevitably goes through software aging due to its characteristics of large and complex middleware and long-time and continuous running. The VMM aging manifests as gradually degrading performance and an increasing failure occurrence rate, due to error conditions that accrue over time and eventually lead the VMM to failure. To counteract the VMM aging, this paper proposes an aging detection and quantification algorithm for Virtual Machine Monitor, which applies Self-organizing Maps (SOM) to capture VMM behaviors from runtime measurement data and takes a neighborhood area density of a winning neuron as an aging quantification metric to detect VMM aging. Results of two experiments injecting different resource leaks on the Xcn platform show that the algorithm has a high true positive rate and a low false positive rate.