System Support for Next Generation Cloud Computing

云计算提供了一种强调按使用付费和资源调配弹性的实用模型。遗憾的是，现有云缺乏广泛的应用程序所需的必要系统级机制，以充分利用云的潜力。这项拟议的研究拓宽了可以利用云计算的应用和服务的范围。我们的第一个研究重点是帮助用户扩大和缩小他们的应用程序的占用空间，以跟踪需求的即时变化。不幸的是，在当前云中实例化新的虚拟机(VM)是一项缓慢的操作(通常需要“几分钟”)，并且新的VM要么作为新的引导程序，要么作为模板VM的副本发起，而不知道当前的应用程序状态。同样，当前的云缺乏在应用需求减弱时简化缩减流程的机制。拟议的研究将创建新的机制，提供必要的扩展和缩减灵活性，使用户能够充分利用云模式的潜力。我们的Second研究推力使在流行数据集上操作的数据密集型计算的高效执行成为可能。虽然虚拟化提供了许多好处(例如，安全性、易管理性以及在程序员配置的定制环境中运行的灵活性)，但它限制了数据重复使用的机会，因为每个虚拟机都必须下载其自己的数据副本。我们的研究将使跨VM的高效文件共享成为可能，并将优化VM放置以最大限度地提高局部性。

Cloud computing provides a utility model that emphasizes pay-per-use and elasticity of provisioning. Unfortunately, existing clouds lack necessary system level mechanisms required by a broad set of applications in order to make full use of the potential of the cloud. The proposed research broadens the set of applications and services that can take advantage of cloud computing. Our first research thrust helps users scale up and down their application's footprint to track instantaneous variations in demand. Unfortunately, instantiating new virtual machines (VM) in current clouds is a slow operation (typically taking "minutes"), and the new VMs originate either as fresh boots or replicas of a template VM, unaware of the current application state. Similarly, current clouds lack mechanisms that simplify the process of scaling down when the demand for an application recedes. The proposed research will create new mechanism that provide the necessary scale-up and scale-down agility to enable users to take full advantage of potential of the cloud model. Our seconds research thrust makes possible the efficient execution of data intensive computations that operate on popular data sets. While virtualization provides many benefits (e.g, security, ease of management, and the flexibility of running in a custom environment configured by the programmer), it limits opportunities for data reuse as each VM has to download its own copy of the data. Our research will make possible efficient file sharing across VMs, and will optimize VM placement to maximize locality.