Software Fault-Tolerance based on Observation and Control of Distributed Programs

为了降低软件故障带来的风险，研究了一种基于前向控制执行的容错方法。在这种方法中，程序在不影响外部环境的情况下以前瞻性模式运行。在forward期间，跟踪足够的信息，以便对其进行故障分析。如果对执行的分析显示没有错误，则提交前瞻性执行，其效果对环境可见。否则，将丢弃前瞻性执行的结果，系统将在受控模式下运行。在这种模式下，系统通过引入额外的同步或更改消息的顺序来执行控制，以避免在forward期间发生的错误。该项目有三个主要组成部分：软件故障检测算法、自动控制算法（如插入控制消息和基于前瞻的消息重新排序）以及支持前瞻控制执行的系统。对于软件故障的检测，以前的算法将被扩展到在存在故障的情况下更一般的计算模型。对于自动控制，项目将重点关注并发程序中的同步故障。最后，在Java中开发了一个支持前瞻控制执行的系统。

To reduce the risks associated with software faults, the research focuses on a fault tolerance method based on controlled execution with lookahead. In this approach, a program is run in a lookahead mode without affecting the external environment. During lookahead, sufficient information is traced so that it can be analyzed for faults. If the analysis of the execution shows that there are no faults, then the lookahead execution is committed and its effects are visible to the environment. Otherwise, the results from the lookahead execution are discarded, and the system is run in a controlled mode. In this mode, the system exercises control by introducing additional synchronization or by changing the order of messages to avoid the fault that occurred during lookahead.There are three main components of the project - algorithms fordetection of software faults, algorithms for automaticcontrol such as insertion of control messages and reordering ofmessages based on lookahead, and a system which supports controlledexecution with lookahead. For detection of software faults, previous algorithms will be extended to work for more general model ofcomputation in the presence of failures. For automatic control, the project will focus on synchronization faults in concurrent programs.Finally, a system in Java that supports controlledexecution with lookahead will be developed.