DAFS Storage for High Performance Computing Using MPI-I/O: Design and Experience

Abstract

A key goal of this effort is to demonstrate and develop heterogeneous and distributed computing technologies that are applicable to DoD and scientific communities, while maintaining and benefiting from industry standards that could be applied to high performance computing. High performance computing systems based on clusters of compute nodes, connected via a high speed interconnect, are becoming popular for large-scale parallel applications, forming a highly scalable infrastructure. Most large-scale scientific applications are highly I/O-centric and have a tremendous need for a similarly scalable file I/O subsystem. DAFS is a well-defined highperformance protocol for file access across a network as well as set of APIs, uDAFS, for user level OS-bypassing application programming, designed to take advantage of RDMA-based transports, such as Virtual Interface Architecture (VIA), InfiniBand Architecture1, and iWARP.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Aug 20, 2004
Accession Number
ADA428747

Entities

People

  • Anthony Skjellum
  • Arkady Kanevsky
  • Peter Corbett
  • Vijay Velusamy

Tags

Communities of Interest

  • Materials and Manufacturing Processes

DTIC Thesaurus Topics

  • Abstracts
  • Computational Processes
  • Computer Access Control
  • Computer Programming
  • Computing System Architectures
  • Computing-Related Activities
  • Data Centers
  • Device Drivers
  • Distributed Computing
  • High Performance Computing
  • Metadata
  • Network Protocols
  • Operating Systems
  • Servers (Computer Hardware)
  • Standards
  • Transport Ships

Fields of Study

  • Computer science

Readers

  • Distributed Systems and Data Platform Development
  • Parallel and Distributed Computing.