Persistent Memory and NUMA

*

One Line Summary

discussion on how persistent memory-aware applications will specify/query NUMA locality

Abstract

It is well understood that application processes benefit from being placed close to their memory on systems with a Non-Uniform Memory Access (NUMA) architecture. There is also benefit (albeit smaller) to locating the application and its resident set on the same NUMA node as storage controllers used for accessing data. This latter point becomes much more important with technology such as persistent memory. The goals of this session include:
- discuss the basic requirements for applications using persistent memory on NUMA systems
- determine the granularity of data locality
- discuss what APIs, new or existing, would be necessary to support persistent memory.

Tags

NUMA, persistent memory

Speaker

  • Jeff Moyer

    Red Hat, Inc.

    Biography

    Jeff has been using Linux since 1994, and has been hacking on the kernel for Red Hat since 2003. He currently focuses on the block layer, the asynchronous I/O and direct I/O code, and persistent memory support.