一句话：提出自适应纠删码温度管理系统 Te-PID，通过预测、过滤与 PID 控制动态调整热冷 EC 格式，提升存储节省并保持访问效率。
云存储系统常采用基于温度的纠删码（EC）格式来平衡数据访问效率与存储成本。系统根据访问频率将数据划分为热 EC 与冷 EC，但数据访问模式的动态性使得有效分类并平衡不同 EC 格式的存储成本与访问效率十分困难。现有静态、基于规则的方法缺乏预测能力，未考虑全局系统状态，也无法动态适应变化环境。为此，作者提出 Te-PID（Temperature Proportional-Integral-Derivative），一种自适应 EC 温度管理系统，将个体访问行为与全局系统模式相结合，包含流量模式预测与存储需求预测模型、减少不必要 EC 格式转换的过滤模块，以及将冷流量比例维持在预设约束内的 PID 控制器。在 Microsoft Azure XTemperature 系统上，包括跨 830 个租户的 30 天真实部署实验表明，Te-PID 相比现有方法将存储节省提升 30.1%，同时保持访问效率并满足系统约束。
Cloud storage systems widely adopt temperature-based erasure coding (EC) formats to balance data access efficiency and storage costs. These systems categorize data into hot and cold EC formats based on access frequency, allocating frequently accessed data to hot EC and less accessed data to cold EC. However, achieving an optimal balance between storage costs with different EC formats and access efficiency through effective data categorization remains a significant challenge due to the dynamic nature of data access patterns. Existing static, rule-based methods lack predictive capability, fail to consider the global system state, and cannot dynamically adapt to changing environments. To address these limitations, we propose Te-PID (Temperature Proportional-Integral-Derivative), an adaptive EC temperature management system that optimizes data storage by integrating individual access behaviors with global system patterns. Te-PID consists of three components: a prediction model to forecast traffic patterns and anticipate storage needs, a filter module to minimize unnecessary EC format transitions, and a PID controller to maintain the cold traffic ratio within predefined constraints, balancing storage costs and performance. Experiments on Microsoft Azure's XTemperature system, including a 30-day real-world deployment across 830 tenants, demonstrate that Te-PID achieves a 30.1% improvement in storage savings compared to existing methods, while maintaining access efficiency and adhering to system constraint.
@inproceedings{paperbot9090,
title = {Te-PID: An Adaptive Erasure Coding Temperature Management System for Optimized Cloud Storage},
author = {Pei Xiao and Lu Wang and Fangkai Yang and Guoqing Geng and Haoran Li and Jeff Zhu and Yu Kang and Yifan Li and Terry Chen and Yue Chen and Saravan Rajmohan and Qi Zhang},
booktitle = {FSE},
year = {2025},
doi = {10.1145/3696630.3728537}
}