Collaborative Research:CNS Core:Medium:Unlinking the (Block)chain: Scalable Byzantine-Tolerant Databases

该项目提出了一种新的架构，用于在互不信任的各方之间实现安全可靠的数据共享。如今，区块链是满足此类需求的首选平台。它们提供了共享的、完全有序的、防篡改的操作日志的抽象，通过由一组计算节点运行的分布式协议实现，每个节点复制日志的整个状态。虽然这些协议的细节因不同的区块链实现而异，但它们都依赖于底层架构，而底层架构受到两个关键限制。首先，它在执行模型上预测正确性和完整性，其中副本按照相同的总顺序顺序处理操作；这很快就会成为性能瓶颈。其次，它使部署复杂化。所有副本存储存储在区块链中的全部数据的需求不仅是资源密集型的，而且可能存在法律问题：例如，GDPR法律严格限制欧洲公民的个人身份数据（如银行或医疗记录）可以驻留在哪里。这个项目通过在一个新的基础上实现区块链抽象来克服这些限制，并且通过这样做，证明了有效的数据处理和信任的去中心化可以齐头并进。它的核心是从先前的数据库研究中获得的一个关键见解——即，有可能实现对完全有序的事务序列的抽象，而不需要所有的操作都按顺序执行。为了利用这种洞察力提供的更大并行性的机会，该项目将开发第一个地理分布式数据库，证明它能够抵御恶意破坏，无论是来自实现数据库的一些计算节点还是来自数据库的客户端。今天，在这样一个数据库中，甚至没有关于正确性的严格规范；该项目将引入这样的正确性规范，基于限制恶意行为者（无论是在实现数据库的副本之间还是在其客户端之间）对正确客户端提交的事务结果的影响。然后，它将开发新的算法技术，尽管有相反的对抗尝试，实现区块链抽象，同时可证明地执行这些正确性条件。为了简化部署，该项目将专注于大幅降低支持区块链所需的复制程度。具体而言，它将改进区块链当前的威胁模型，将确保不丢失数据所需的复制程度与保证区块链执行完整性所需的复制程度解耦；这样就可以放松所有副本存储区块链的整个状态的要求，从而更容易实现法规遵从性。相互不信任的各方之间更紧密的集成可能对各种应用程序（包括医疗保健、金融服务和供应链管理）产生变革性影响。数据库是支持这种集成的自然候选对象，但是目前缺乏一种方法来表达和有效地执行某些参与者恶意行为时的正确性。这项工作构建了概念框架，利用一种新颖的架构，使其更容易遵守数据使用规则，并在工作原型中实例化它，这些原型将被严格评估其性能和健壮性。理论和工程之间的协同作用对这个项目的成功至关重要，对下一代系统工程师的教育也至关重要：他们将不得不在支撑当今大多数行业的无处不在的大规模分布式系统中协商性能、容错和信任问题。拟议的讲座和项目材料将使学生准备应用有原则的方法来构建可信赖的分布式系统。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

This project proposes a new architecture for enabling safe and secure data sharing among mutually distrustful parties. Blockchains are today the platform of choice for such needs. They offer the appealing abstraction of a shared, totally-ordered, tamper-resistant log of operations, implemented via a distributed protocol run by a collection of computing nodes, each replicating the entire state of the log. While the details of these protocols vary for different blockchain implementations, they all rely on an underlying architecture that suffers from two key limitations. First, it predicates correctness and integrity on an execution model where replicas process operations sequentially, according to the same total order; this quickly becomes a performance bottleneck. Second, it complicates deployment. The need for all replicas store the full body of data stored in the blockchain is not only resource intensive, but can be legally problematic: GDPR laws, for instance, severely restrict where personally identifiable data (such as banking or medical records) of European citizens can reside. This project overcomes these limitations by implementing the blockchain abstraction on a novel foundation, and, in so doing, demonstrates that efficient data processing and the decentralization of trust can go hand-in-hand. At its core is a key insight gained from prior research in databases – namely, that is possible to implement the abstraction of a totally ordered sequence of transactions without requiring all of their operations to be actually executed sequentially. To leverage the opportunity for greater parallelism offered by this insight, this project will develop the first geo-distributed database provably resilient to malicious sabotage, whether from some of the computing nodes implementing the database or from clients of the database. Today, there exists no rigorous specification of what correctness would even mean in such a database; this project will introduce such correctness specification, based on the ethos of bounding the influence that malicious actors (whether among the replicas implementing the database or among its clients) can have on the outcome of transactions submitted by correct clients. It will then develop new algorithmic techniques that, despite adversarial attempts to the contrary, implement the blockchain abstraction while provably enforcing these correctness conditions. To simplify deployment, this project will focus on drastically reducing the degree of replication necessary to support blockchains. Specifically, it will refine the blockchain’s current threat model to decouple the degree of replication needed to ensure that no data will be lost from what is required to guarantee the integrity of the blockchain execution; this will make it possible to relax the requirement that all replicas store the entire state of the blockchain, and thus lead to easier regulatory compliance. Tighter integration among mutually untrusted parties can be transformative for a variety of applications, including healthcare, financial services, and supply chain management. Databases are natural candidates for supporting this integration, but currently lack a way to express and efficiently enforce correctness when some actors behave maliciously. This work builds that conceptual framework, leveraging a novel architecture that makes it easier to comply with data use regulations, and instantiates it in working prototypes that will be rigorously evaluated for performance and robustness. The synergy between theory and engineering critical for the success of this project is also crucial to the education of the next generation of system engineers: they will have to negotiate issues of performance, fault-tolerance, and trust in the ubiquitous large-scale distributed systems that underpin most industries today. Proposed lecture and project materials will prepare students to apply a principled approach to building trustworthy distributed systems.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.