Enabling a highly-scalable, cloud-based microservices architecture

本研究的目标是设计、构建和评估机制，使高度可扩展的、基于云的微服务架构能够支持未来的、基于云的企业应用。在这里，企业应用程序将是由许多服务组成的微服务体系结构的实例。未来的操作规模将测量数十万个服务实例。云计算客户希望他们的应用程序有无限的计算需求，但是云计算提供的计算资源是有限的。数据中心在空间和运营成本方面达到其物理极限的情况并不少见。今天，云提供商为满足高价值客户而应用的实用解决方案是为他们提供专用的、非常昂贵的、功能强大的硬件，而其他硬件资源仍然在其他客户之间大量共享。然而，客户的计算需求，无论是高价值的还是其他的，往往被无意或有意地夸大了，这样昂贵的物理资源有时会闲置，而同一数据中心中其他邻近的资源则被推到了极限。此外，并非所有托管的应用程序都期望在任何时候都经历峰值负载，因此可能会出现某种程度的物理资源过度供应。在我们的方法中，服务将被封装在容器中。容器是轻量级虚拟机。与权重较大的虚拟机相比，它们提供隔离、快速按需启动和实时迁移，并且占用资源较少。我们的方法将利用新颖的混合扩展技术，动态地上下扩展容器的计算、网络和存储资源，以满足客户对其应用程序的性能期望，同时最大限度地减少云提供商的物理资源使用。

The goals of this research are to design, build and evaluate mechanisms that enable a highly-scalable, cloud-based microservices architecture to support future, cloud-based enterprise applications. Here, enterprise applications will be instances of microservices architectures that are comprised of many services. The future scale of operation will measure hundreds of thousands of service instances.Cloud customers desire infinite computing needs for their applications but cloud provides have finite computing resources to work with. It is not uncommon that data centers are reaching their physical limitations in terms of space and operating cost. Today, the pragmatic solution applied by cloud providers to satisfy high-value customers is to provide them with dedicated, highly expensive, and powerful hardware, while other hardware resources remain heavily shared among other customers. However, computing needs of customers, whether high-value or other, are often over-stated, either inadvertently or on purpose, such that expensive physical resources may sit around idle at times, while other close by resources in the same data center are pushed to their limits. Also, not all hosted applications are expected to experience peak load at all times, such that some degree of physical resource over-provisioning is possible.In our approach services will be encapsulated in containers. Containers are lightweight virtual machines. They offer isolation, fast on demand start-up, and live migration, at a low resource footprint, as compared to their heavier weight virtual machine counterparts. Our approach will leverage novel hybrid scaling techniques that dynamically scale the compute, networking and storage resources of containers up and down in order to meet the performance expectations expressed by customers on their applications, while minimizing physical resource usage for the cloud provider.