Containers versus Virtual Machines
This article gives an overview of containers for those who know what a Virtual Machine (VM) is. VMs are implemented by products such as VMware, Xen, Hyper-V, Virtual Box, Parallels Desktop etc. If you do not have experience with those or similar products, better read What are containers whitepaper.
Containers are very similar to VMs in a sense that they also let one to partition one physical computer into multiple small isolated partitions (called VMs or containers). The difference is in technique used for such partitioning.
Some of the major differences between VMs and containers, as well as their consequences, are outlined below.
Single kernel concept
Xen, KVM, VMware and other hypervisor-based products provide an ability to have multiple instances of virtual hardware (called VMs – Virtual Machines) on a single piece of real hardware. On top of that virtual hardware one can run any Operating System, so it's possible to run multiple different OSs on one single server. Each VM runs full software stack (including an OS kernel).
In contrast, OpenVZ uses a single-kernel approach. There is only one single OS kernel running, and on top of that there are multiple isolated instances of user-space programs. This approach is more lightweight than VM. The consequences are:
- Waiving the need to run multiple OS kernels leads to higher density of containers (compared to VMs)
- Software stack that lies in between an application and the hardware is much thinner, this means higher performance of containers (compared to VMs)
File system
From file system point of view, a container is just a chroot() environment. In other words, a container file system root is merely a directory on the host system (usually /vz/root/$CTID/, under which one can find usual directories like /etc, /lib, /bin etc.). The consequences are:
- there is no need for a separate block device, hard drive partition or filesystem-in-a-file setup
- host system administrator can see all the containers' files
- containers backup/restore is trivial
- there is no I/O overhead (for VMs it can be as high as 1.5x to 3x, especially for small requests)
- mass deployment is easy