Wait analysis of virtual machines using host kernel tracing

An agent-less method to understand virtual machines (VMs) behavior its evolution during the VM life-cycle is an essential task for IaaS provider. It allows the IaaS provider to better scale the VMs resources by properly allocating the physical resources. On the other hand, because of privacy, security, ease of deployment and execution overhead issues, the method presented limits its data collection to the physical host level, without internal access to the VMs. We propose a host-based, precise method to recover wait states for the virtual CPUs (vCPUs) of a given VM. The Wait Analysis Algorithm (W2A) computes the state of vCPUs through the host kernel trace. The state of vCPUs is displayed in an interactive trace viewer (TraceCompass) for further inspection. Our proposed VM trace analysis algorithm has been open-sourced for further enhancements and to the benefit of other developers. Our new technique is being evaluated with representative workloads, generated by different benchmarking tools. These approaches are based on host hypervisor tracing, which brings a lower overhead (around 0.03%) as compared to other approaches.

Mots clés

virtual machine; performance analysis; wait analysis; elasticity