CAREER: Cost-Effective Application Deployments Spanning Multiple Cloud Services

云计算服务的使用-托管基础设施的服务，第三方可以在其上运行应用程序和存储数据-在过去几年中迅速增长。然而，尽管云服务的数量在不断增加，但目前大多数应用程序都部署在单一的云服务上。这种现状有几个缺点-应用程序提供的可用性和性能受到部署应用程序的云服务的限制，应用程序提供商为实现性能目标而承担的成本受到该服务提供的资源的类型和定价的限制。本研究项目将通过规划和管理跨多个云服务的应用程序部署，使分布式应用程序能够以更低的成本获得更好的可用性和性能。认识到应用程序的不同组件具有不同的特点，该项目将通过将应用程序组件的异构性与不同云服务提供的资源多样性相匹配，实现经济高效的多云部署。该项目中的研究涉及三项主要任务-1)开发算法，以便在众多候选配置中经济高效地选择用于跨云服务部署应用程序；2)设计和实现分布式测量和监控框架，以便随着云服务和互联网随时间的变化而动态重新配置应用程序的部署；以及3)开发工具和技术，使为部署在单一云服务上而实施的遗留应用程序能够轻松移植到多云部署。这项研究将涉及大量的系统构建和应用于在真实云服务上部署流行应用程序的拟议技术的试验。广泛影响：尽管云服务越来越受欢迎，但对其使用仍存在一些担忧。作为这项研究的一部分开发的算法和软件系统可能会改变应用程序提供商使用云服务的方式，从而缓解对可用性和性能的担忧。该项目还包含一个重要的教育部分，引入了新的研究生和本科课程，并举行了公开讲座，以提高小型和非营利性企业对云服务的认识。

The use of cloud computing services---services that host infrastructure on which third-parties can run applications and store data---has skyrocketed in the last few years. However, despite the growing number of cloud services, most applications are deployed today on a single cloud service. This status quo has severaldrawbacks---availability and performance offered by an application are limitedby the cloud service on which the application is deployed, and cost borne byan application provider to meet performance goals is limited by the types andpricing of resources offered by that service.This research project will enable distributed applications to obtain betteravailability and performance at lower cost by planning and managing application deployments that span multiple cloud services. Recognizing the fact that differentcomponents of an application have different characteristics, this project willenable cost-effective multi-cloud deployments by matching the heterogeneity ofan application's components with the diversity of resources offered bydifferent cloud services. The research in this project involves three maintasks---1) the development of algorithms to choose cost-effectively among the myriad candidate configurations for deploying an application across cloud services, 2) the design and implementation of a distributed measurement and monitoring framework to dynamically reconfigure an application's deployment asthe state of cloud services and of the Internet change over time, and 3) thedevelopment of tools and techniques that enable legacy applications implemented for deployment on a single cloud service to be easily ported to multi-cloud deployments. The research will involve significantsystems-building and experimentation with the proposed techniques applied todeployments of popular applications on real cloud services.Broader Impact:In spite of the increasing popularity of cloud services, there remain severalconcerns about their use. The algorithms and software systems developed aspart of this research may transform the ways in which application providersuse cloud services, thus easing concerns with respect to availability andperformance. The project also has a significant education component with theintroduction of new graduate and undergraduate classes, and public talks toincrease awareness about cloud services among small and non-profit businesses.