SaTC: CORE: Small: Efficient Hardware-Aware and Hardware-Enabled Algorithms for Secure In-Memory Databases

许多大数据工作负载都托管在云上，以便于共享和低成本。其中许多工作负载都要处理敏感数据，例如电子健康记录。云基础设施容易受到不受信任的运营商的攻击，这些运营商拥有对计算机的物理访问权限。这些攻击可以采取多种形式，并可能危及隐私或完整性。为了保证敏感数据的安全并确保共享网络基础设施的成本效益，在不造成性能、能量或成本方面的重大损失的情况下挫败这些攻击是至关重要的。目前，上述攻击是通过不经意的RAM(ORAM)和完整性树来挫败的。这些解决方案会产生大量带宽和容量开销，并可能使性能降低1-2个数量级。这些解决方案还在不同的工作负载上表现出不同的行为，即在评估算法的安全性时不能忽视系统、硬件和工作负载的影响。为了解决这些问题，PI正在设计新的安全硬件，并利用商品安全硬件来设计新的安全系统。PI正在开发需要硬件/软件协同设计的新的安全算法，因为它们的行为是工作负载局部性属性和硬件约束的函数，例如，由定制存储模块实现的ORAM的分布式实现。这些新算法正在集成到可扩展的内存数据库系统中，这些数据库系统运行在一群最先进的安全节点上。因此，该项目试图保护敏感应用程序的网络基础设施，将众所周知的方法(ORAM和完整性树)转移到实际领域。警察督察也在继续其外展和教育工作。

Many big-data workloads are hosted on the cloud to facilitate sharing and low cost. Many of these workloads deal with sensitive data, e.g., electronic health records. Cloud infrastructures are vulnerable to attacks from untrusted operators with physical access to the computers. These attacks can take many forms and may compromise privacy or integrity. To guarantee the security of sensitive data and ensure the cost-effectiveness of shared cyberinfrastructure, it is vital that these attacks be thwarted while not incurring a significant penalty in terms of performance, energy, or cost.Currently, the above attacks are thwarted with oblivious RAM (ORAM) and integrity trees. These solutions incur significant bandwidth and capacity overheads and can degrade performance by 1-2 orders of magnitude. These solutions also show different behaviors on different workloads, i.e., one cannot ignore system, hardware, and workload effects when evaluating algorithms for security. To address these problems, the PIs are designing novel secure hardware, and leveraging commodity secure hardware to design novel secure systems. The PIs are developing new secure algorithms that require hardware/software co-design because their behavior is a function of workload locality properties and hardware constraints, e.g., a distributed implementation of ORAM enabled by custom memory modules. These new algorithms are being integrated into scalable in-memory database systems that run on a cluster of state-of-the-art secure nodes. The project is thus attempting to secure cyberinfrastructure for sensitive applications, moving well-known approaches (ORAM and integrity trees) into the practical realm. The PIs are also continuing their outreach and education efforts.