Research on Method to Construct Scalable Server Systems with Fault-Tolerance

在这项研究中，我们提出了一个新的模型和机制，分布式资源管理，其中有容错，效率，可扩展性和迁移能力，我们已经开发了一个实现的模型和机制。在SSS-PC操作系统的研究和开发中，我们提出了在多台PC机上建立一个虚拟化UNIX域的方法，并提出了一种可扩展和可迁移的进程机制。该研究将引导SSS-PC的研究和开发进入下一阶段。到2006财政年度结束时，取得了以下成绩。可以在多台PC机上的一个SSS-PC系统中建立UNIX域的任意片段。此外，还开发了SSS-PC系统中三个TCP协议栈协同工作的容错协议栈。使用该协议栈的容错文件服务器系统在2007财年上半年进行了实验性实施。一个文件服务器程序在三台PC机上冗余执行，该程序使用容错协议栈与外部客户端通信。在这种情况下，它被证明是保持提供服务，即使其中一台PC机的文件服务器程序运行故障。然而，澄清了冗余执行中的名称空间重叠通过该实验性实现成为一个问题。为了从根本上解决这一问题，必须将每个冗余执行程序视为名称空间上的独立存在。因此，执行方法被切换为在另一个虚拟化UNIX域上分别实现冗余执行的每个程序执行的方法，并且在2007财年下半年增强了UNIX域的虚拟化机制。

In this study, we have made a new model and mechanism for distributed resource management in which there are fault-tolerance, efficiency, scalability, and migration-ability and we have developed an implementation of the model and the mechanism. In the research and development of the SSS-PC operating system, we had proposed a methodology to make one virtualized UNIX domain over multiple PC machines, and invented a scalable and migratable process mechanism. This study would lead the SSS-PCs research and development to the next stage. The following was achieved by the end of fiscal year 2006. The arbitrary pieces of UNIX domains could be established in one SSS-PC system over multiple PC machines. In addition, the fault-tolerant protocol stack communicated with one other party as three TCP protocol stacks cooperated in the SSS-PC system was developed. The fault-tolerant file server system using this protocol stack was experimentally implemented in the first half of fiscal year 2007. One file server program was redundantly executed with three PC machines, and the program used the fault-tolerant protocol stack to communicate external client terminals. In this situation, it was proven to keep providing service even if one of PC machines where the file server program was running broke down. However, it was clarified that the overlap of the name space in redundant execution became a problem through this experimental implementation. For the fundamental solution of the problem, it was concluded that it was necessary to treat each program of redundant execution as independent existence on the name space. Therefore, the execution method was switched to the method to achieve each program execution of redundant execution respectively on another virtualized UNIX domain, and virtualization mechanisms for UNIX domain was enhanced in the latter half of fiscal year 2007.