RESEARCH INITIATION AWARD: Investigating A Hybrid Function and Data Shipping Parallel Programming Environment

9308879 Carter这个项目的目标是为分布式内存多处理器开发一种新颖的编程环境，为数据传输和函数传输编程范例提供有效的支持。这是通过集成分布式共享内存（DSM）系统和远程过程调用（RPC）系统来实现的，前者提供透明的数据传递，后者支持分布式函数传递。这种集成解决了两种范式中最严重的问题。DSM使进程能够在仅通过消息传递进行通信的处理器之间共享数据，但是使用Munin DSM系统的经验表明，要求所有通信都通过共享数据进行通常是低效的。相反，在传统的RPC系统中，很难传递使用任意指针的复杂数据结构，因为客户机和服务器驻留在不同的地址空间中。这严重限制了编程模型。支持共享地址空间允许将任意复杂的数据结构作为RPC的一部分传递。初步结果表明，数据和功能传递的结合可以显著提高DSM的性能，提高RPC的可编程性。该项目还涉及开发简单的性能监控和可视化工具，以简化编程这样一个系统的任务。＊＊＊

9308879 Carter The objective of this project is to develop a novel programming environment for distributed memory multiprocessors that provides efficient support for both the data shipping and function shipping programming paradigms. This is accomplished by integrating a distributed shared memory (DSM) system, which provides transparent data shipping, with a remote procedure call (RPC) system, which supports distributed function shipping. This integration addresses the most serious problems of both paradigms. DSM enables processes to share data across processors that only communicate via message passing, but experience with the Munin DSM system indicates that it is often inefficient to require all communications to occur through shared data. Conversely, in conventional RPC systems, it is difficult to pass complex data structures that use arbitrary pointers because clients and servers reside within distinct address spaces. This seriously restricts the programming model. Supporting a shared address space allows arbitrarily complex data structures to be passed as part of an RPC,. Preliminary results indicate that the combination of data and function shipping can significantly improve the performance of DSM and improve the programmability of RPC. The project also involves the development of simple performance monitoring and visualization tools to ease the task of programming such a system. ***