[论文解读] Brief Announcement: Update Consistency in Partitionable Systems

Matthieu Perrin, Achour Mostéfaoui+1 more|arXiv (Cornell University)|Jan 9, 2015

Distributed systems and fault tolerance参考文献 2被引用 4

一句话总结

本文提出了更新一致性（UC），这是一种针对可分割分布式系统的新型一致性准则，确保副本收敛到由更新的全序排列导出的状态，同时保持每个进程的顺序。UC 严格强于最终一致性，且具有普遍可实现性，通过更新的全序排列和顺序规范，使分布式对象的实现具备模块化、可组合性及形式化可验证性。

ABSTRACT

Data replication is essential to ensure reliability, availability and fault-tolerance of massive distributed applications over large scale systems such as the Internet. However, these systems are prone to partitioning, which by Brewer's CAP theorem [1] makes it impossible to use a strong consistency criterion like atomicity. Eventual consistency [2] guaranties that all replicas eventually converge to a common state when the participants stop updating. However, it fails to fully specify shared objects and requires additional non-intuitive and error-prone distributed specification techniques, that must take into account all possible concurrent histories of updates to specify this common state [3]. This approach, that can lead to specifications as complicated as the implementations themselves, is limited by a more serious issue. The concurrent specification of objects uses the notion of concurrent events. In message-passing systems, two events are concurrent if they are enforced by different processes and each process enforced its event before it received the notification message from the other process. In other words, the notion of concurrency depends on the implementation of the object, not on its specification. Consequently, the final user may not know if two events are concurrent without explicitly tracking the messages exchanged by the processes. A specification should be independent of the system on which it is implemented. We believe that an object should be totally specified by two facets: its abstract data type, that characterizes its sequential executions, and a consistency criterion, that defines how it is supposed to behave in a distributed environment. Not only sequential specification helps repeal the problem of intention, it also allows to use the well studied and understood notions of languages and automata. This makes possible to apply all the tools developed for sequential systems, from their simple definition using structures and classes to the most advanced techniques like model checking and formal verification. Eventual consistency (EC) imposes no constraint on the convergent state, that very few depends on the sequential specification. For example, an implementation that ignores all the updates is eventually consistent, as all replicas converge to the initial state. We propose a new consistency criterion, update consistency (UC), in which the convergent state must be obtained by a total ordering of the updates, that contains the sequential order of each

研究动机与目标

• 解决最终一致性在可分割系统中指定共享对象时的局限性。
• 提供一种与实现细节（如消息传递并发性）无关的一致性准则。
• 通过将规范与系统级并发性解耦，实现形式化验证与模型检测。
• 提出一种普遍、可组合的一致性模型，支持分布式对象的模块化编程。
• 通过依赖顺序规范和全序更新，克服并发规范的复杂性。

提出的方法

• 将更新一致性（UC）定义为一种一致性准则，要求所有副本收敛到由尊重每个进程本地顺序的更新全序导出的状态。
• 结合 Lamport 时间戳与进程 ID，构建更新的全序，确保副本间的一致性。
• 通过按全序重放更新，事后重构历史，实现一致的状态收敛。
• 将对象规范解耦为抽象数据类型（顺序行为）与一致性准则（分布式行为），支持形式化验证。
• 通过证明 UC 的普遍性，表明任何对象均可在具有任意崩溃故障的可分割系统中，利用全序机制实现。
• 利用关注点分离，支持分布式规范的自动编译与可组合性。

实验结果

研究问题

• RQ1 能否定义一种与消息传递并发性及系统级实现细节无关的一致性准则？
• RQ2 如何仅通过顺序语义与一致性准则来指定和验证分布式对象？
• RQ3 是否存在一种强于最终一致性、能保证收敛到正确状态的一致性准则？
• RQ4 能否设计一种通用构造方法，在具有任意崩溃故障的可分割系统中实现任意对象的更新一致性？
• RQ5 更新一致性能否在分布式系统中实现可组合性，从而替代复杂的组合式实现技术？

主要发现

• 更新一致性（UC）严格强于最终一致性（EC），确保收敛状态由尊重每个进程本地顺序的更新全序导出。
• 仅当历史可线性化，使得删除不以无效方式出现在插入之后时，历史才为更新一致，如图 1(c) 所示。
• 本文证明了在可分割系统中，即使存在任意崩溃故障，每个对象都存在更新一致的实现，确立了 UC 作为普遍一致性准则。
• 使用 Lamport 时间戳与进程 ID 可实现更新的确定性全序，可用于在副本间重构一致状态。
• 更新一致性支持规范的自动编译，而非实现，从而通过顺序规范实现可组合性与形式化验证。
• 该方法将规范与系统级并发性解耦，使得可应用模型检测与形式化验证技术于分布式对象。