UIUC paper on how kernel-space loads determine the best course of action for business workflows deployed as either Virtual Machines or Kubernetes containers.

The original document is linked below. This research has been co-authored with Martin Becerra (UIUC) and Hyunkyung Lee (UIUC).

Modern apps are designed to be easily deployed in various on-prem and cloud environments. The two most popular methods are Virtual Machines and containers. We performed a large scale experiment of the most popular business workloads to determine which deployment provides the best ROI on your infrastructure cost. In the process, we’ve identified time spent in kernel-space as a high correlation factor of the performance output. This is not surprising: container based deployments share the kernel of their host node, while VMs have their own virtualized OS.