Realizing a Adaptable Distributed Object-Oriented Computing Environment by Using Reconfigurable Object

在大规模的开放分布式计算环境中，运行时环境，如性能和可靠性，都在不断变化。如果用户希望系统在运行时环境变化的情况下仍保持一定程度的性能和可靠性，则系统应该适应运行时环境的变化。这些变化在系统或应用程序的生命周期中是不可预测的。在本研究中，我们提出了一种可重构对象，该对象可以根据运行环境的变化进行自我重构。可重构对象由一些功能元对象以及用户定义的元对象组成。当可重构对象检测到运行时环境的变化时，它会将一些功能元对象交换到另一个功能元对象，以适应这种变化。为了方便可重构对象交换其元对象，我们在元对象之间采用了松散耦合的通信。因此，由可重构对象组成的分布式系统具有适应其运行环境的能力。通过实现可重构对象的原型和环境变化检测机制，我们的想法是可以实现的。此外，我们还展示了该系统的四个应用领域。第一种是自适应容错，它使系统适应底层系统的可靠性和用户要求的可靠性水平的变化。第二种是自适应远程访问，它根据系统性能的变化，解决了远程数据访问的成本和质量之间的权衡问题。第三种是在互斥群成员发生变化时的法定人数重构。最后一种是自适应消息排序协议，即使系统性能发生变化，也能选择性价比高的协议。通过这些实例，我们证实了可重构对象有助于实现具有成本效益、健壮性和可靠性的分布式系统。

In large scale open distributed computing environments, runtime environments, such as performance and reliability, are continuously changing. If a user wants the system to preserve some level of performance and reliability in spite of change of runtime environments, the system should adapt itself to change of runtime environments. These changes are not predictable in the life time of systems or applications.In this research, we proposed a reconfigurable object that can reconfigure itself according to change of the runtime environment. The reconfigurable object consists of some functional meta-objects as well as a user defined meta-object. When the reconfigurable object detects a change of runtime environment, it exchanges some functional meta-objects to another one to adapt itself to the change. To make it easy for a reconfigurable object to exchange its meta-objects, we adopted loosely-coupled communication between meta-objects. Thus the distributed system that is composed of the reconfigurable objects has an ability to adapt to its runtime environment.We confirmed that our ideas can be realized by implementing the proto-type of reconfigurable objects and an environment change detection mechanism. Moreover, we showed following four application areas of the system. The first one is adaptable fault tolerance, which adapts the system to change of the underlying system's reliability and user required level of reliability. The second one is adaptable remote access, which solves the trade-off between cost and quality of data on accessing remote data, according to change of system performance. The third one is a quorum reconfiguration on the occasion of change of mutual exclusion group members. The last one is adaptable message ordering protocols, which selects the cost effective protocols even if system performance changes. Through these examples, we confirmed that reconfigurable objects are useful to realize a cost effective, robust and reliable distributed system.