Supercomputing

• HPC Support
• Get a New OSC Account
• Request Additional
   Resources
• Request a Classroom
   Project
• FAQ
• Computing
  Environments
• Available Hardware
• Training
• Accounts
• Available Software
• Statewide Software
• Manuals
• Consult
• Notices

Related Links

• Statewide Users Group
• Blue Collar Computing
• Ralph Regula School
  of Computational
  Science
• BALE
• Interface Lab
• Cluster Ohio Project

Parallel I/O Techniques

Description:

 

This course is intended as an introduction to parallel input/output techniques. As the sophistication of parallel applications increases, I/O is now often the greatest performance bottleneck, especially for applications dealing with very large datasets. Furthermore, modern parallel architectures are increasingly designed with parallelism in their I/O subsystems as well as in their processor/memory interconnect network.

 

The course will consist of four major sections. The first section will introduce the concept of parallel I/O and describe how it can be implemented in hardware. The second section will describe several very simple approaches to parallel I/O, as well as why these approaches may not be suitable for very large datasets. The third section will discuss the MPI-2 I/O interface, which is the de facto standard parallel I/O API. The final section will describe HDF5, a high level scientific data format and library built on top of the MPI-2 I/O API.

 

Prerequisites and Requirements:

 

Attendees should be familiar with parallel programming using MPI in either C or Fortran.

 

Target Audience:

 

Users of parallel systems such as the OSC IA32 cluster, or OSC/SGI IA64 cluster.

 

Handouts:

 

November 2004, PDF, by Troy Baer (troy@osc.edu)

May 2003, PDF, by Troy Baer (troy@osc.edu)

March 2002, PDF, by Troy Baer (troy@osc.edu)