This paper introduces and compares different schemes\udfor assignment of replicated services in Jini - an object oriented middleware architecture for network-centric computing. Each client in Jini has to be assigned a service\udselected from the pool of available services, which have\udjoined the Jini federation and registered with the lookup\udservice. Both early and delayed assignments are considered\udas basic options in our evaluation. The information\udfor the system load can be collected at four different levels of detail in order to be used in the process of assignment decisions. In our analysis, we concentrate on the scenario where the requests for service generated by the clients follow independent user-initiated or machine-initiated transactions. The performance evaluation of the assignment schemes follows the queuing systems methodology. The comparisons are done with regard to the mean residence time of the clients in the system as well as the control overhead imposed by the assignment schemes. A case study of the scheme using the lowest information level proves the effectiveness, applicability and limitations of the delayed assignment in comparison to the early one. These results are a first step towards developing a methodology for building large-scale applications for Jini-based distributed systems.\ud
展开▼
机译:本文介绍并比较了Jini中用于复制服务分配的不同方案\ ud,这是一种面向对象的中间件体系结构,用于以网络为中心的计算。必须从可用服务池中为Jini中的每个客户端分配一个服务\ udselected,这些服务已\ u加入Jini联合身份并向lookup \ udservice注册。在我们的评估中,早期任务和延迟任务都被视为\ udas基本选项。可以在四个不同的详细级别上收集系统负载的信息,以便在分配决策过程中使用。在我们的分析中,我们专注于由客户生成的服务请求遵循独立的用户启动或机器启动的事务的场景。分配方案的性能评估遵循排队系统方法。关于客户端在系统中的平均停留时间以及分配方案所施加的控制开销进行比较。通过使用最低信息级别的方案的案例研究,证明了与早期分配方案相比,延迟分配方案的有效性,适用性和局限性。这些结果是开发开发用于基于Jini的分布式系统的大规模应用程序的方法的第一步。\ ud
展开▼