Collaboration Summit 2014

This talk introduces the NUMA technology in the Linux kernel first showing how a typical Enterprise system may use NUMA features and introducing the basic operations to configure and view the NUMA configuration of the current system. Building on that then unusual use cases are explored that allow creative use of the NUMA features. For example high end HP systems have a large shared memory node in addition to the local memory of each node that we find on typical NUMA systems. Embedded systems may use a NUMA node to segment memory in order to place performance critical objects into memory that is backed by SRAM or flash. These features are also used in Haswell/Broadwell processors with a special configurable "l4" caches and also in the upcoming new generation of Xeon Phi processsors by Intel. Given enough time we can explore some other exotic uses of NUMA.

Developers and system administrators that want to make more advanced use of the NUMA features in the Linux Kernel will find this presentation interesting.