Impact of Virtualization and Containers on Application Performance and Energy Consumption
TimeTuesday, July 246:30pm - 8:30pm
DescriptionIt is well-known that applications run faster when run on bare-metal; however, virtualization is omnipresent and is endemic to cloud systems. For some applications, e.g. legacy code or applications which spend much of the time idle, virtual machines are a good choice and one needn’t really worry about performance costs; however, there are certainly situations in which one would want to know about the performance impact of virtualization and containers. Here, we compare performance penalties on a stand-alone server while using KVM (kernel-based virtual machine) virtualization and containers on Docker and Singularity for creating isolated environments for SPEC ACCEL and SPEC OMP2012 benchmarks suites. We then repeat this procedure in a virtual machine and in containers (KVM, Docker, Singularity) on a stand-alone server and on the Jetstream cloud system. For a subset of tests on the stand-alone server, we also measure energy virtualization impacts energy efficiency as previous work has shown that virtualization increases energy consumption as it reduces computational performance.