NYI: Operating Systems for Multiscale Computers

VLSI技术的进步使得分布式系统(例如，工作站集群)适合于并行计算。然而，今天的软件限制了分布式系统的功能。这项研究的目标是发现和评估(1)位置和工作分配的自动管理，(2)故障隔离，(3)高性能通信，和(4)在成本效益高的分布式系统中的公平共享的技术。该方法基于共享对象、基于外内核的操作系统、用户级通信单元和群组调度。该项目建立了一个实验性的分布式系统，提供与定制设计的超级计算机相当的性能，同时保持故障隔离和公平共享。

Advances in VLSI technology have made distributed systems (e.g., clusters of workstations) suitable for parallel computing. Today's software, however, limits the functionality of distributed systems. The goal of this research is to discover and evaluate techniques for (1) automatic management of locality and work distribution, (2) fault isolation, (3) high-performance communication, and (4) fair sharing in a cost-effective distributed system. The approach is based on shared objects, an exokernel-based operating systems, a user-level communication unit, and gang scheduling. The project builds an experimental distributed system that delivers performance comparable to custom-designed supercomputers, while maintaining fault isolation and fair sharing.