Recent Advances in the Linux Kernel Resource Management

The Linux kernel is the solid base for a modern multitask/multiuser (and now even multicontainer) operating system. The kernel runs many of concurrent tasks fighting for the limited set of resources, so it has to manage/share those resources in a fair and controlled manner. The main resources are: CPU, RAM (kernel and user memory), swap, disk space, disk I/O, and network bandwidth.

The talk outlines existing resource control mechanisms (mainly ulimit) and their shortcomings (including, but not limited to the inability to control process groups rather than individual processes, and the inability to account/control kernel memory used on behalf of the userspace processes). Then it goes on to describe what containers are and why group-based resource management is of high importance for containers. The recent additions to the Linux kernel – control groups (a generic mechanism for grouping tasks and assigning different resource controllers to those groups) and memory controller (a way to account and limit various memory-related resources) are explained in details, with some interesting implementation details and a look into the features that are to be added.

Intended audience: developers, system administrators, anyone interested in containers virtualization and/or resource management.