System Support for Scalable Distributed Services

随着互联网络计算的普及，公共服务将被大量广泛分布的客户端访问。因此，这种服务的可伸缩实现必须处理（1）大量客户端，以及（2）在广泛分布的节点中常见的高延迟。 减少服务器负载和通信成本的一种方法是在频繁访问服务的客户端组附近创建服务的多个实例。此外，如果可能，资源丰富的客户端可以缓存服务状态，以避免常见访问的通信延迟。服务状态复制和缓存都导致服务状态的多个副本，并且在服务状态可以被更新时导致跨副本的一致性维护的问题（例如，以支持跨客户端的交互）。这个项目的重点是在大规模系统的不同应用程序集的一致性水平，是适当的表征。 此外，将开发可扩展协议，以提供对服务状态的读写访问，而不需要与可能具有服务状态副本的所有节点进行通信。 一个基于分布式对象的中间件将实现支持可伸缩的服务，它将采用对象缓存和复制，以提高服务访问性能和可伸缩性。该中间件将用于开发和评估通过POTS和电缆网络等各种通信介质到达的针对家庭的服务。

As internetworked computing becomes pervasive, common services will be accessed by a large number of widely distributed clients. Thus, scalable implementations of such services must deal with (1) large number of clients, and (2) high latencies that will be common among widely distributed nodes. One approach for reducing server load and communication costs is to create multiple instances of a service close to client groups that frequently access it. Also, when possible, resource rich clients can cache service state to avoid communication latency for common accesses. Service state replication and caching both lead to multiple copies of the service state and lead to the problem of consistency maintenance across copies when service state can be updated (e.g., to support interactions across clients). The focus of this project is on the characterizations of consistency levels that are appropriate in large scale systems with diverse sets of applications. In addition, scalable protocols will be developed to provide both read and write access to service state without requiring communication with all nodes that may have copies of the service state. A distributed object based middleware will be implemented to support scalable service and it will employ object caching and replication to enhance service access performance and scalability. This middleware will be used in the development and evaluation of services targeted to the home that are reached via a variety of communication media such as POTS and cable networks.