CSR--PDOS: Improving System Reliability via Delta Execution

许多应用程序需要高可靠性和可用性。不幸的是，软件故障和管理重新配置错误极大地降低了系统的可靠性和可用性，每种故障占系统故障的26-30%。为了提高软件可靠性和减少管理错误，在软件生命周期的不同阶段使用了许多可靠性保证技术，包括软件测试、在线软件补丁验证、基于部分复制的故障检测和恢复以及管理重新配置的在线验证。有趣的是，上述所有可靠性保证任务都有一个共同特征：多次几乎冗余执行（MARE）。换句话说，它们都执行同一软件的多个版本/副本，每次执行只在代码段、输入或配置方面与其他执行略有不同。通常用于执行MARE的现有解决方案是在相同或不同的机器上依次（如软件测试）或同时（如在线补丁和重新配置验证）完全执行这些几乎冗余的运行。这些方法固有的低效率以两种方式之一表现出来：大开销或资源和人力的大量浪费都会对可靠性的验证时间和成本产生负面影响。为了有效地支持用于执行各种可靠性保证任务的MARE，我们提出了一种称为增量执行的新方法，并提出了一个全面的基础设施来支持它。为了解决delta执行中出现的研究挑战，本研究将多个层面的创新紧密集成在一起：操作系统、动态和静态编译、编程语言支持和各种可靠性保证。

Many applications require high reliability and availability. Unfortunately, software failures and administrative reconfiguration errors greatly reduce system reliability and availability, each contributing to 26-30% of system failures. To improve software reliability and reduce administrative errors, many reliability assurance techniques have been used at different stages of the software life cycle, including software testing, online software patch validation, partial replication-based fault detection and recovery, and online validation of administrative reconfigurations. Interestingly, all of the above reliability assurance tasks share a common characteristic: multiple almost-redundant executions (MARE). In other words, they all execute multiple versions/copies of the same software, each execution differing from the others only slightly in code segment, input, or configuration. An existing solution commonly used to perform MARE is to fully execute these almost-redundant runs sequentially (as done for software testing) or simultaneously (as done for online patch and reconfiguration validation) on the same or different machines. The inherent inefficiency of these approaches materializes in one of two ways large overhead or large waste of resources and human effort both negatively impacting the validation time and cost for reliability.To efficiently support the MARE used in performing various reliability assurance tasks, we propose a novel approach called delta execution and a comprehensive infrastructure to support it. To address the research challenges arising in delta execution, the proposed research tightly integrates innovations from multiple layers: operating sytsem, dynamic and static compilation, programming language support and various reliability assurance.