MANTA: Machine Learning Augmented Tiering Advisor

2026-09-30

AI Summary

This work addresses the responsiveness limitations of existing heuristic memory tiering strategies under bursty workloads, which hinder efficient utilization of fast storage tiers. We propose a machine learning-based adaptive page migration approach. Specifically, we first design ChOMP, an offline optimizer that profiles performance bottlenecks and establishes benchmarks. Runtime access features are then extracted to train a lightweight predictive model that dynamically perceives page hotness, guiding migration decisions between CXL and Optane memory tiers. Integrating this model into the ARMS framework enables a paradigm shift from static heuristics to predictive tiering. Experimental evaluations demonstrate that the proposed method achieves average speedups of 1.08–1.12× in Linux CXL environments and up to 1.69× in Optane scenarios, with peak improvements reaching 5.6× for individual workloads.

Abstract

Memory tiering has been used to expand memory capacity, particularly in datacenters, by combining fast DRAM with slower tiers, including CXL-attached memory. Its effectiveness depends on keeping useful pages in the fast tier, but existing heuristic policies can lag behind changing hot sets in phased or bursty workloads. To explore these limitations, we introduce ChOMP, a scalable offline optimizer that minimizes placement and bandwidth-sensitive migration costs. We then develop a trace-driven simulator that uses this reference to identify performance opportunities for online policies. Motivated by these results, MANTA predicts future page usefulness from runtime access features and integrates a lightweight learned model into ARMS. Across eight workloads on emulated CXL, MANTA achieves geometric-mean speedups over ARMS of 1.12$\times$ and 1.08$\times$ at 4~GB of fast memory on Linux 6.2 and 6.18, respectively; across six Optane workloads, it achieves 1.69$\times$. On individual workloads, MANTA is up to 1.25$\times$ faster than ARMS with emulated CXL on Linux 6.2, 1.21$\times$ on Linux 6.18, and 5.6$\times$ with Optane.

Problem

• memory tiering
• page placement
• heuristic policies
• CXL-attached memory
• workload dynamics

Innovation

• Memory Tiering
• Machine Learning
• Offline Optimizer
• Page Usefulness Prediction
• CXL Memory