多虚拟机系统HDD-SSD混合磁盘I/O调度及其性能优化研究

由于资源复用与服务聚合，磁盘I/O系统已成为多虚拟机系统的性能瓶颈。为提高虚拟化系统的磁盘I/O性能，同时兼顾系统成本，目前采用传统机械式硬盘(HDD)和固态硬盘(SSD)混合系统，其中SSD充当高速缓存或作为部分应用的专属存储，大容量HDD作为主要存储设备。由于HDD和SSD在性能、操作方式上的异构性，传统的面向单一的非虚拟化环境下的SSD或HDD的磁盘调度策略，无法适用于多虚拟机间磁盘I/O高度竞争的HDD-SSD混合磁盘。本课题根据虚拟化环境下HDD-SSD混合磁盘系统调度机理、负载模式及性能约束上的重大变化，研究虚拟机环境下HDD-SSD混合磁盘系统性能模型、性能、服务质量、公平性等指标联合约束的磁盘I/O调度策略、I/O性能优化与能耗优化方法。通过进行细粒度的HDD-SSD联合磁盘I/O调度和性能优化，提高多虚拟机系统的磁盘I/O性能和服务质量，同时进一步降低系统能耗。

The Disk I/O subsystem becomes as the bottleneck of massive virtualized systems due to resource multiplexing and services consolidation. Hybrid HDD-SSD storage is a promising solution to address this challenge. In Hybrid HDD-SSD systems, SSDs serve as Caches for traditional HDDs or dedicated storage for exclusive applications, while HDDs are base storage devices tradeofffing among the I/O performance gains and total cost of ownership. However, traditional disk I/O scheduling strategies fail in such virtualized Hybrid HDD-SSD systems because of the heterogeneity and difference in performance and I/O operations in HDDs and SSDs respectively. In this project we plan to investigate the disk I/O performance model, performance,fairness, and quality of service aware disk I/O scheduling strategies, I/O performance optimization, and energy optimization in virtualized Hybrid HDD-SSD systems. Through fine-grained co-scheduling of HDDs and SSDs, the I/O performance will be improved for mass virtual machines and their energy consumptions are also optimized.