面向多租户的虚拟数据中心可靠性设计

With the increasing popularity of commercial cloud computing applications, there is an increasingly urgent need for data centers, which are the key building blocks of cloud computing, to provide tenants with guaranteed reliability services. This project focuses on investigating the reliability design of virtual data centers operating in a multi-tenant environment with frequent equipment failure and large demand variations. The goal is to provide tenants with predictable and explicit reliability guarantee at minimum cost. To meet various reliability requirements and overcome the deficiency of limited physical layer reliability choices, we propose a reliability enhancement scheme on the virtualization layer. The key idea is to allocate additional computing and networking resources for each tenant's virtual data center demand. The proposed reliability enhancement scheme will provide the capability to adaptively meet various tenant reliability requirements without the need to change existing reliability configurations on the physical layer. To avoid the problem of fluctuations in reliability performance in the presence of highly-varied demands, we propose an agile reliability scheme. The key idea is to reserve a proper amount of elastic computing and networking resources to tolerant a wide range of demand variation scenarios and leverage the asynchronous nature in demand variations among tenants to propose an over-subscription concept to further improve resource utilization efficiency. The advantages of the agile reliability scheme include fast and efficient response to large demand variations and maintaining the stability and modularity of reliability. To tolerate large-scale failures and demand-variations, we propose a collision-avoidance progressive reconfiguration approach. We will also establish an experimental platform based on real data centers to test and validate our proposed approaches and ensure that they are practically applicable.

随着云计算商业应用的普及，作为云计算核心的数据中心向其租户提供可靠性保障服务显得日益迫切。本项目重点研究失效频繁发生、需求动态变化和多租户并存环境下虚拟数据中心的可靠性设计。目标是用最小的代价使租户的可靠性得到可预期的明确保证。为满足租户的多样化可靠性要求以及克服物理层可靠性服务单一的不足，提出虚拟层可靠增强机制，其核心是在虚拟层添加冗余计算资源和网络资源以增强可靠性；优点是无需改变物理层的可靠性配置就能灵活满足租户的区分可靠性要求。为避免需求动态变化下可靠性波动的问题，提出敏捷可靠机制，其核心是为每个租户预留弹性计算和网络资源以容忍该租户大部分变化场景并利用租户间需求变化的非同步性超额订购资源。敏捷可靠的优点是高效、快速应对需求的动态变化并保证可靠性的稳定和隔离。用冲突避免多级重配置机制解决大规模失效或变化下多点快速重配置问题。并搭建实验平台对所提的方法进行测试和验证，使其符合实际应用。

实现各种云计算服务的数据中心虚拟化技术已经引起了许多关注。随着数据中心虚拟化的普及，有效地将虚拟数据中心（VDC）映射到物理数据中心上，使得VDC的生存性能够被保证显得愈来愈重要。在本项目中，我们研究了在设备故障频繁和大量需求变化的多租户环境下虚拟数据中心的可靠性设计问题，目标是以最低成本为租户提供可预测和明确的可靠性保证。我们提出了虚拟层增强方法来定制生存性，根据VDC可靠性要求在虚拟层定制具有冗余节点和链路的VDC请求。仿真结果表明，所提出的虚拟化层增强概念对共享资源和提高资源利用率具有重要影响。为了避免在需求高度变化情况下可靠性波动的问题，我们提出了一种基于相关的虚拟机整合方案。 其关键思想是利用租户间需求变化的异步性，进一步提高了资源利用效率。为了容忍大规模失效和需求变化，我们还研究了冲突避免的重配置策略，提出了基于改进的串行迁移策略和并行迁移策略的混合迁移策略。此外，我们开发了排队模型来量化性能指标。我们通过数学分析评估了所提出的迁移策略的性能，数值结果表明我们提出的策略优于现有方法。我们还建立了一个基于Openstack的实验VDC平台来测试和验证我们提出的方法。