Synapse
⌘+K
Synapse
PulseExploreJournal ClubResearchersJournals
Instagram
HomeJournal ClubExplore
April 25, 2026ACM SIGMOD Record

Lineage-based Reuse and Memory Management for Multi-backend ML Systems

View Full Paper
Ask AI
Bookmark
Share

Authors

APArnab PhaniMBMatthias Boehm

Discussion

Loading...

Member takes

Overview

Randomized trial shows holistic memory management enhances multi-backend reuse, suggesting efficiency improvements in ML systems.

Key Points

  • This paper aims to address efficient reuse and memory management in multi-backend machine learning systems.
  • Introduced the MEMPHIS framework for multi-backend reuse and memory management.
  • Developed a lineage-based reuse cache for managing data across different backends.
  • Implemented cache management policies for asynchronous execution and workload-aware operations.
  • Achieved up to 9.6x performance improvement compared to existing ML systems across various tasks.
  • Demonstrated effective management of memory allocation overheads and bandwidth constraints.

Cite This Study

Phani et al. (2026) studied this question.

synapsesocial.com/papers/69ec5a2588ba6daa22dabb9chttps://doi.org/10.1145/3810900.3810916
View Full Paper
Ask AI
Bookmark
Share

Also Consider

Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context:

  1. 1Technical Perspective on 'MEMPHIS: Holistic Lineage-based Reuse and Memory Management for Multi-backend ML Systems'2026
  2. 2LIMA: Fine-grained Lineage Tracing and Reuse in Machine Learning Systems2021 · 30 citations
  3. 3eLLM: Elastic Memory Management Framework for Efficient LLM Serving2025
  4. 4MemVault: A Three-Layer Hierarchical Memory Management System for Cost-Optimized LLM Applications2026
  5. 5Accelerating LLM Inference via Dynamic KV Cache Placement in Heterogeneous Memory System2025