Non-invasive estimation of cloud applications performance via hypervisor’s operating systems counters

The adoption of cloud computing environments as the infrastructure of choice for computing services is growing rapidly, due to features such as scalability and pay-per-use. As a result, more pressure is put on cloud providers, which manage the underlying computing platform, to maintain the Quality of Experience of application users within acceptable levels. However, the mapping of high-level application metrics, such as response time, to low-level infrastructure metrics, such as utilization rate of resources, is a non-trivial task. Many works present monitoring of processor, memory, and network utilization. Nevertheless, the monitoring of these resources can be intrusive to the system that provides the service. This paper presents a non-invasive approach for estimating the response time of cloud applications through the mapping of Quality of Service metrics to operating system counters at the hypervisor level. We developed a model that estimates the response time of real-time applications based on Linux Operating Systems counters that presented an accuracy of 94% in our evaluation