CSR: Small: Cost-Aware Cloud Profiling, Prediction, and Provisioning as a Service

云计算通过实现对大规模计算资源的即时、按需访问，有可能改变计算实践。但大规模的云计算很容易代价高昂。可扩展的成本感知型云基础设施管理和配置(SCRMP)项目旨在开发新的云访问方法，以降低复杂性和成本，并提高云资源的使用效率。该项目将在三个领域进行创新：概况分析、预测和供应。其新的基于机器学习的评测技术旨在根据可比较和相关的实例类的派生、探索性评测技术和历史使用分析，以不同的精度水平预测不同云资源集的应用程序性能。其基于集成的市场预测模型将允许轻松比较和合并许多现有的云市场预测模型，以便可以利用它们的集体优势来预测成本，从而将成本、价格风险和实例吊销的可能性降至最低。最后，SCRIMP的总体配置模型将结合应用程序配置文件和市场预测模型，以实现自动化、经济高效、基于策略的云配置，以及在所产生的动态配置环境中高效地放置和迁移工作负载。SCRIMP将通过简化对按需云计算的访问并提高研究人员使用云基础设施的效率，推动跨科学计算的使用，特别是在较小的机构内。通过降低许多用户的科学计算成本和复杂性，SCRIMP将能够更有效地使用云信用(无论是来自云提供商还是资助机构)，使研究人员在没有专用计算基础设施或专业知识的情况下更容易访问云计算，并允许研究人员和学生在更大的数据集上以更高的分辨率进行日益复杂的分析。SCRIMP还将与教育直接相关，使教育工作者能够以低成本提供访问大型资源库的机会，并保证性能。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。

Cloud computing has the potential to transform computational practice by enabling immediate, on-demand access to large-scale computing resources. But large-scale cloud computing can easily be costly. The Scalable Cost-Aware Cloud Infrastructure Management and Provisioning (SCRIMP) project aims to develop new cloud access methods that will reduce the complexity and cost and improve the efficiency of using cloud resources. The project will innovate in three areas: profiling, prediction, and provisioning. Its new machine learning-based profiling techniques aim to predict application performance, at different levels of accuracy, across a diverse set of cloud resources, based upon derivation of comparable and related instance classes, explorative profiling techniques, and analysis of historical usage. Its ensemble-based market prediction models will allow the many existing cloud market prediction models to be easily compared and then combined so that their collective strengths can be used to predict costs with the aim of minimizing cost, price risk, and likelihood of instance revocation. Finally, its overarching provisioning model will combine application profiles and market prediction models to enable automated, cost-efficient, policy-based cloud provisioning as well as efficient placement and migration of workload within the resulting dynamically provisioned environment.SCRIMP will advance the use of computation across the sciences, particularly within smaller institutions, by simplifying access to on-demand cloud computing and improving the efficiency with which researchers make use of cloud infrastructure. By lowering scientific computing costs and complexity for many users, SCRIMP will enable more efficient use of cloud credits (whether from cloud providers or funding agencies), democratize access to cloud computing by researchers without dedicated computing infrastructure or expertise, and allow researchers and students to conduct increasingly complex analytics, on larger datasets, and at higher resolution. SCRIMP will also be directly relevant in education, allowing educators to provide access to large resource pools at low cost with guaranteed performance.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.