CNS Core: Small: RUI: Optimal and Efficient Resource Allocation in Policy-Driven Data Centers: A Network Flow Approach

新兴的数据和通信密集型应用，如在线视频流、社交媒体、健康和医疗应用、包括高能物理和生物信息学的科学应用，以及教育应用(如快速增长的大规模在线公开课)，都由大型云数据中心提供支持。然而，由于管理数据中心资源的复杂性越来越高，支撑这一基础设施的压力越来越大。包括亚马逊云存储和微软Skype在内的领先科技公司以及Gmail和Whatsapp等流行移动应用的云服务频繁中断就证明了这一点。为了应对这一挑战，该项目将为策略驱动的数据中心(PDDC)创建一个优化且高效的资源分配框架，以集成的方式管理云用户应用程序和云资源(即服务器、网络和电源)。该项目的目标是将计算、数据和中间盒(MBS)这三个PDDC的构建块集成到一个框架中，以实现最佳的云资源管理。PDDC中的各种重要问题，包括虚拟机(VM)的迁移和放置、负载平衡、流优先级和容错，都可以使用网络流技术来解决，这些技术提供了最佳和高效的资源分配解决方案。特别是，该项目确定了一系列自适应地协调计算、数据和MB的新策略保存问题，并发明了一套策略保存算法，在高效消耗云资源的同时满足不同的云策略。建议的技术包括在PDDC中放置、迁移、复制和流量工程计算、数据和MB。该项目将把结果与基于整数线性规划(ILP)的解决方案进行比较，并将该方法扩展到多目标优化问题。预期成果是PDDC的基本理论、架构、算法和协议，以及提供长期政策保护云服务的原型。该奖项反映了NSF的法定使命，并通过使用基金会的智力优势和更广泛的影响审查标准进行评估，被认为值得支持。

Emerging data- and communication-intensive applications such as online video streaming, social media, health and medical applications, scientific applications including high-energy physics and bioinformatics, and educational applications (such as the rapidly-growing Massive Open Online Courses) are all enabled by large cloud data centers. However, this underpinning infrastructure is increasingly stressed by the growing complexities of managing data center resources. This is evidenced by the frequent outages of cloud services from leading tech companies, including Amazon Cloud Storage and Microsoft Skype, and popular mobile apps such as Gmail and Whatsapp. To address this challenge, this project will create an optimal and efficient resource allocation framework for policy driven data centers (PDDCs), to manage cloud user applications and cloud resources (i.e., servers, networks, and power) in an integrated fashion. The goal of this project is to integrate compute, data, and middleboxes (MBs), three building blocks of PDDCs, into one framework to achieve optimal cloud resource management. A variety of important problems in PDDCs, including virtual machine (VM) migration and placement, load balancing, flow priority and fault tolerance can all be solved using network flow techniques that provide optimal and efficient resource allocation solutions. In particular, the project identifies a series of new policy-preserving problems that adaptively coordinate compute, data, and MBs, and invents a suite of policy-preserving algorithms that satisfy diverse cloud policies while consuming cloud resources efficiently. The proposed techniques include placing, migrating, replicating, and traffic engineering compute, data, and MBs in the PDDC. The project will compare results with integer linear programming (ILP)-based solutions and extend the approach to multi-objective optimization problems. Expected outcomes are fundamental theories, architectures, algorithms, and protocols for the PDDCs, and prototypes that provide long term policy-preserving cloud services.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.