软件定义的云计算基础理论与方法的研究进展

Achievements of Study on the Theory and Method of Software Defined Cloud Computing

当前云计算用户体验与资源利用率之间的矛盾已成为阻碍其快速发展的瓶颈。共享云系统仍然存在尾延迟、低效率、高干扰三大应用问题。亚马逊、谷歌等企业已对上层软件栈开展了大量优化,但仍无法有效解决问题。因此工业界呼吁通过底层硬件创新、软硬件协同方式来保障服务质量、提高资源利用率。学术界也已开始探索从底层硬件到上层应用的全系统栈协同设计方案。本项目提出面向第三代云计算发展方向的低熵云计算理论框架,通过研究标签化冯·诺依曼体系结构、容器化存储引擎、通信局部化高阶网络、时空共享资源管理、混合负载测试基准SDCBench等,实现在共享云环境下同时满足用户体验与资源利用率,较国际主流云平台降低尾延迟1个数量级,提升资源利用率1倍。

The current contradiction between user experience and resource utilization has become a bottleneck to cloud computing.Cloud systems have three application problems:tail latency,low efficiency,and high interference.Amazon,Google and other companies have carried out a lot of optimizations on the upper software stack,but still cannot effectively solve the problem.Therefore,the industry calls for the underlying hardware innovation and software-hardware co-design to protect application QoS and improve resource utilization.The academic community has also begun to explore a system-wide stack collaborative design from the underlying hardware to the upper application.This project proposes a low-entropy cloud computing theoretical framework for the third-generation cloud computing.We further propose labeled von Neumann architecture,containerized storage engine,communication localized high-order network,space-time shared resource management,mixed-load datacenter benchmark SDCBench.Based on these techniques,we will achieve both user experience and resource utilization in a shared cloud environment,reduce the tail latency by one order of magnitude compared to the international mainstream cloud platform,and improve resource utilization by one time.