Explore the performance characteristics and optimal usage of Intel's Optane DIMM in this 29-minute conference talk from FAST '20. Gain insights into the behavior of scalable persistent memory, learn best practices for maximizing device performance, and discover how to reoptimize application-level software for persistent memory. Delve into topics such as latency, bandwidth, buffer size, store instructions, and NUMA access considerations. Understand the implications of this new technology for programming models, file systems, and applications designed to exploit nonvolatile DIMMs.
An Empirical Guide to the Behavior and Use of Scalable Persistent Memory
Overview
Syllabus
Intro
Outline
Background: Optane in the Machine
Background: Optane Internals
Background: Optane Interleaving
Basics: Our Approach
Test Platform
Basics: Latency
Basics: Bandwidth
Avoid small random accesses
Lessons: Optane Buffer Size
Use ntstores for large writes
Lessons: Store instructions
Lessons: Contention at Optane Buffer
Lesson: Contention at IMC
Limit threads accessing one NVDIMM
Avoid mixed and multi-threaded NUMA accesses
Conclusion
Taught by
USENIX
