Skip to content

Persistent Memory I/O Primitives (A. van Renen, DaMoN 2019) #5

@benclmnt

Description

@benclmnt

Guidelines for effective usage of PMem

  • Algos should be designed to fit data on PMem blocks (256 byte) rather than single cache lines (64 byte)
  • Use streaming ops or stores followed by clwb, especially for data written to the same cache line, (e.g. array-like structures with size field or a global counter for time-stamping)
  • Using too many threads can lead to reduced performance
  • PMem read & write bandwidth is lower than DRAM. Prefer DRAM for performance-critical code.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions