An Early Exploration of Deep-Learning-Driven Prefetching for Far Memory

NeurIPS 2025 Workshop MLForSys Submission48 Authors

Keywords: Machine learning for systems, Memory prefetching, Far memory, Inference optimization, Memory access prediction, Linux kernel

TL;DR: Memix is a deep-learning-based Linux far-memory system that minimizes on-demand accesses through efficient DL-guided prefetching, achieving up to 42% performance improvement over the state of the art.

Abstract:

Far-memory systems, where applications store less-active data in more energy-efficient memory media, are increasingly adopted by datacenters.
However, applications are bottlenecked by on-demand data fetching from far- to local-memory.
We present $\textbf{\textit{Memix}}$,
a far-memory system that embodies a deep learning–system co-design for efficient and accurate prefetching, minimizing on-demand far-memory accesses.
One key observation is that memory accesses are shaped by both application semantics and runtime context, providing an opportunity to optimize each independently.
Preliminary evaluation of Memix on data-intensive workloads shows that it outperforms the state-of-the-art far-memory system by up to 42%.

Submission Number: 48