Implementation of Fast Data Transfer with Global Flush Broadcasts

点击翻译按钮获取中文摘要

9523815 Camp Flush communication channels generalize the more conventional asynchronous one-to-one communication paradigms. A flush channel offers the implementor of a distributed application the flexibility of specifying a message delivery order apropos of the demands of the application. This stands in marked contrast to the rigid FIFO (first-in-first-out) delivery order of communication specified by a virtual circuit in TCP/IP or specified by a virtual channel in ATM communication. The more restrictive the delivery order, the less concurrency possible to exploit in message transmissions over a multi-path system. Flush channels allow the user to define the delivery of a message to be after the delivery of all messages previously transmitted, or before the delivery of all messages subsequently transmitted, or both. The delivery order between the sending site and the destination site, created by the user, is a partial order. The results from previous work are encouraging. Flush communication channels can reduce the mean delivery time of messages substantially, regardless of the underlying physical structure of the network. That is, an implementor of a distributed application that uses an F-channel can reduce communication costs whether the communication occurs on routes established for virtual circuit or datagram in TCP/IP or UDP/IP or on routes established for the connectionless service class D provided by the ATM Adaptation Layer. This possible reduction in mean delivery time of messages may be critical in meeting real-time requirements of the application. However, all previous results only consider interprocess communication (IPC) for one-to-one communication. The goal of this research is to investigate the implications of implementation and performance when we extend F-channel communication to many-to-many communication. ***