Platform-Agnostic Lightweight Deep Learning for Garbage Collection...
Talkat 12th USENIX Workshop on Hot Topics in Storage and File Systems (HotStorage '20) · 2020-07-13..07-14·
summary
A platform-agnostic GC scheduler called GC Tutor uses a lightweight DNN classifier to predict the idle time between the last and next I/O request and schedules flash block reclamation during predicted idle windows. It adds a meta-learning online adaptation step so models trained offline adapt within minutes to new workloads that caused fixed models to drop below 30% accuracy (proxy traces). Real-workload evaluation shows GC-induced latency falls by 82.4% on average versus threshold rules.
AI-written summary · about
chapters (5)
related talks
15mHow can machine learning assist to improve JVM performance?
13mGNNAdvisor: An Adaptive and Efficient Runtime System for GNN Acceleration on GPUs
15mEffectively Scheduling Computational Graphs of Deep Neural Networks toward Their...
11mLitePred: Transferable and Scalable Latency Prediction for Hardware-Aware Neural...
13mLeveraging AI for Faster Storage Access: A Graph-Neural-Network-Based Prefetcher
10mLightning Talk: Profiling and Memory Debugging Tools for Distributed ML Workloads on GPUs
18mTierTrain: Proactive Memory Tiering for CPU-Based DNN Training
21mA heuristic algorithm for periodic memory allocation with little fragmentation to train(…)