A Study of Virtualization Overheads

Containers or OS-based virtualization have seen a recent resurgence in deployment.
Containers have low memory footprint and start-up time but provide weaker security
isolation than Virtual Machines (VMs). Incapability to load kernel modules and support
multiple OS, and platform-dependence limits the functionality of containers. On
the other hand, VMs or hardware-based virtualization are platform-independent and
are more secure, but have higher overheads. A data centre operator chooses among
these two virtualization technologies---VMs and containers---when setting up guests
on cloud infrastructure.
Density
and
Latency
are two critical factors for a data centre
operator because they determine the efficiency of cloud computing. Therefore, this
thesis contributes updated density and latency measurements of KVM VMs and Linux
Containers with a recent kernel version and best practices. This work also studies the
memory footprint of KVM VMs and Linux Containers. In addition, it identifies three
ways to improve the density of KVM VMs by lowering the memory footprint: improving existing memory
deduplication techniques, removing unused devices emulated by
QEMU, and removing unused pages from the guest address space.