AF: Medium: A Fair Prescription for Partial Synchrony

AF：中：部分同步的公平处方

• 批准号: 0964696
• 负责人: Jennifer Welch
• 金额: $ 73.77万
• 依托单位: Texas A&M Engineering Experiment Station
• 依托单位国家: 美国
• 项目类别: Continuing Grant
• 财政年份: 2010
• 资助国家: 美国
• 起止时间: 2010-06-01 至 2015-05-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0964696&HistoricalAwards=false
• 关键词: AF; Medium; Fair; Prescription; Partial

Partial synchrony refers to computing environments where timing bounds exist for communication delays and processing speeds, but knowledge of these bounds may be limited or unknown. Although such bounds are only implicit, they can still be used to advantage for building reliable distributed systems in the presence of message loss and crashed processors. The central problem however, is how timing parameters for partial synchrony should be modeled, measured, and denominated.Since its inception some 25 years ago, the prevailing paradigm of chronometry-based models has invoked real-time as a formal basis for modeling temporal bounds on message delays and process speeds. Unfortunately, such models are intrinsically limited (and even flawed) as descriptions of empirical distributed systems, because the characteristic property of partial synchrony is not chronometric timeliness, but rather chronological fairness.This project is developing a fundamental theory of partial synchrony based on the notion of chronological fairness. The key technical innovation is that system failures can be detected in executions that are unfair, rather than untimely. This approach leads to greater generality because many untimely executions are still fair. Theoretical results codify a hierarchy of fairness properties, including technical limitations, expressivity, and reducibility to oracular models. Practical results extract fairness models from empirical systems. An integrated critique of chronometric models helps to initiate a broader shift in research trends to focus on fairness-based paradigms of partial synchrony.

部分同步是指存在通信延迟和处理速度的定时界限的计算环境，但这些界限的知识可能是有限的或未知的。虽然这样的界限只是隐式的，但它们仍然可以用于在存在消息丢失和处理器崩溃的情况下构建可靠的分布式系统。然而，中心问题是如何定时参数的部分同步应建模，测量和national.Since成立约25年前，基于计时模型的流行范式调用实时作为一个正式的基础上建模的时间界限的消息延迟和处理速度。不幸的是，这些模型作为经验分布系统的描述本身是有限的（甚至是有缺陷的），因为部分同步的特征属性不是时间上的时间性，而是时间上的公平性。本项目正在发展一个基于时间公平性概念的部分同步的基本理论。关键的技术创新是，系统故障可以在不公平而不是不及时的执行中检测到。这种方法导致了更大的普遍性，因为许多不合时宜的处决仍然是公平的。理论结果编纂了公平属性的层次结构，包括技术限制，表现力，并还原为神谕模型。实践结果从经验系统中提取公平模型。对计时模型的综合批判有助于引发更广泛的研究趋势转变，将重点放在基于公平的部分同步范式上。