Collaborative Research: High Performance Parallel Processing: Fault-Tolerant Computing on a Network of Workstations

点击翻译按钮获取中文摘要

This project deals with both theoretical and prototyping research in parallel processing to harness the power of workstation clusters. The primary techniques will based on aggressive scheduling, evasive memory and dispersed data management. The formal work expands the previous results of the PIS in asynchronous computing in several directions: (1) Extend the models to models to fine-grained programs. (2) Permit the inclusion of architecture specific characteristics into theoretical frameworks. (3) Techniques to distinguish inconsistencies in memory states as viewed by different processors resulting from time-outs, interrupts and failures (or indefinite postponement of a computation). (4) Techniques to allow non-deterministic executions. (5) Model input/output operations of fault-tolerant programs. Thee experimental part of the project will build a scaled version to execute parallel programs on clusters of workstations. The prototype will address issues related to the eager scheduling of threads, the implementation of evasive memory and data management. The prototype will attempt to evaluate the effectiveness of the environment in terms of efficiency. This project is conducted jointly by P. Dasgupta of Arizona State University and Z. Kedem, of New York University.