High Performance Virtual Machines
Abstract
The primary goal of the High Performance Virtual Machines (HPVM) project was to reduce the effort required to build a high performance cluster and distributed applications by leveraging the investments and understanding of scalable parallel systems. The approach to reducing the programming effort required to build high performance distributed applications was to insulate the program with a uniform, portable abstraction - a High Performance Virtual Machine - with predictable, high performance characteristics. Success was achieved by delivering a large fraction of the underlying hardware performance, visualizing resources to provide portability and reduce the application building effort and delivering predictable, high performance computing. HPVM has produced several major software releases involving high performance communication libraries, complete cluster software systems, and improved versions of those systems on a variety of hardware and software platforms. These systems have been downloaded and deployed at top research universities, major computer companies and national research laboratories. Clusters of commodity systems connected by high-speed networks are an important computing element and serve as an important model for future computing environments. Technologies which effectively exploit distributed computational resources can tap this "cluster pool" to deliver high performance computing, dramatically increasing the computational power available for both high performance computing and high performance distributed applications.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 01, 2002
- Accession Number
- ADA408475
Entities
People
- Andrew A. Chien
Organizations
- University of Illinois Urbana–Champaign