AdaptLSTM improves cloud workload forecasting with less compute cost
AdaptLSTM: Efficient Adaptive Online Learning for Cloud Workload Forecasting under Distribution Drift
Machine Learning
Summary
Cloud services need to predict how much work they'll get to adjust resources like computers and storage. But sudden changes, like viral events or product launches, make old prediction models inaccurate quickly. The authors created AdaptLSTM, a method that spots when these changes happen and updates predictions only when needed. This makes the predictions more accurate while using much less computing power than updating all the time. AdaptLSTM works well with different types of prediction models and was tested on real cloud data.
What this means in practice
- •For cloud infrastructure teams: Reduce resource provisioning errors by using efficient adaptive workload predictions to handle sudden usage changes.
- •For real-time data engineers: Integrate adaptive online learning techniques to maintain prediction accuracy under shifting data patterns with lower compute cost.
Authors
Xinhua Miao, Bowei Yang, Zhengong Cai
Abstract
Accurate workload forecasting is critical for elastic resource provisioning in web-scale cloud services, where distribution shifts driven by viral content, product launches, and user behavior degrade offline-trained models rapidly. Naive online learning recovers accuracy but incurs prohibitive per-step compute cost. We propose AdaptLSTM, an adaptive online framework that detects drift via validation-calibrated thresholds and applies selective, targeted updates. On the Alibaba Machine Trace, AdaptLSTM recovers 54\% of Naive Online's improvement at 20\% cost ($2.7\times$ efficiency, $p=0.002$ over 10 seeds). On the more volatile Container Trace, it achieves 96\% at 20\% cost ($4.8\times$ efficiency, $+75\%$ MAE reduction over Static). Unlike classical drift detectors (ADWIN, DDM, Page-Hinkley) which fail to trigger on regression-scale error streams, AdaptLSTM fires 42 times over 301 steps and outperforms matched-budget baselines. Wall-clock profiling shows $1.33\times$ throughput gain and 45\% update-time reduction. The framework is model-agnostic: identical Pareto patterns hold for LSTM, GRU, and Transformer backbones.