SaTC: CORE: Small: Program Analysis and Transformations for Secure Computation on the Cloud

云服务提供商提供广泛的数据存储和计算产品。客户越来越多地将数据和计算外包给第三方云提供商。然而，当客户将他们的数据上传到云提供商时，他们放弃了数据机密性。正如最近发生的无数事件所证明的那样，保护数据和计算的安全是一个极其重要的问题。该项目探讨客户是否可以利用包括计算在内的廉价、高效和便捷的云服务，同时在一定程度上保护数据机密性并保持效率。该项目构建了新的程序分析，以减轻数据转换的影响，并使用这些方法将程序转换为安全的计算。这项研究利用了SecureMR之前的工作，SecureMR是一个系统，可以自动将MapReduce程序转换为在公共云上处理加密数据的程序。该项目解决了SecureMR的一个关键问题，即单个受信任的客户端必然会限制高度并行的MapReduce程序的可伸缩性。该项目设想在两个云系统之间划分工作的程序分析和转换，并探索在构建保证数据机密性的模型和分析方面的关键挑战，从而使任何一个系统都无法重建原始明文输入数据。该项目开发了基于两方计算(2PC)的SecureMR-2PC，将输入数据拆分成两个云之间的共享，以避免聚合数据的泄漏，实现可证明的隐私保障。此外，该项目还开发了新的程序分析技术，用于分析共享协议以提高多方计算的效率。该项目构建了一套全面的MapReduce程序基准套件，以促进云上安全计算的研究。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 service providers offer a wide range of data storage and computation products. Customers increasingly outsource data and computation to third-party cloud providers. However, when customers upload their data to the cloud provider, they relinquish data confidentiality. As countless recent events attest, securing data and computation is a problem of tremendous importance. This project explores whether customers can take advantage of inexpensive, efficient, and convenient cloud services, including computation, while preserving to some extent data confidentiality, and retaining efficiency. This project builds new program analyses that mitigate the impact of data conversion, and transform programs to compute securely using these methods. This research leverages prior work on SecureMR, a system that automatically transforms a MapReduce program into one that works over encrypted data on a public cloud. This project addresses a key issue with SecureMR, namely, that a single trusted client is bound to limit scalability of highly parallel MapReduce programs. This project envisions program analyses and transformations that split work between two cloud systems, and explores key challenges in building models and analyses that guarantee confidentiality of data such that neither system can reconstruct the original plaintext input data. The project develops SecureMR-2PC, based on 2-party computation (2PC), which splits input data into shares among two clouds, to avoid leakage of aggregate data and achieve provable privacy guarantees. Additionally, the project develops novel program analysis techniques that reason about sharing protocols to improve efficiency of multi-party computation. The project builds a comprehensive benchmark suite of MapReduce programs to facilitate research on secure computation on the cloud.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.