SysOptic: A Fine-Grained Monitoring System for Virtual Machines Based on PMU

Modern cloud data centers indicate the frequent existence of complex failure manifestation. Failures have become the norm, and not the exception. This is a key challenge since assumptions that underpin designing reliable systems are monitoring systems status and detecting anomaly at runtime. Performance Monitoring Unit on CPU (PMU) can obtain fine-grained monitoring data by adopting interrupt sampling method based on hardware events. However, profilers in virtual machines fail to receive PMU relevant information directly due to the limited capacity of PMU virtualization. In this paper, we present a fine-grained monitoring system SysOptic based on PMU virtualization. First, we propose a method to expose PMU information PMU and ensure the visibility ofsuch information at virtual machine level. Second, to maximize the PMU reusability, SysOptic supports the PMU sharing and simultaneous monitoring among multiple virtual machines. Furthermore, we also describe how to synchronize interrupts on physical machines to virtual machines by using injecting interrupts. Experimental results show that with the aid of SysOptic, profiler tools in virtual machines manage to perceive the existence of PMU and collect the monitoring data. The additional overhead incurred by SysOptic is at most 9.8%.