Translation techniques for distributed-shared memory programming models

PDF Version Also Available for Download.

Description

The high performance computing community has experienced an explosive improvement in distributed-shared memory hardware. Driven by increasing real-world problem complexity, this explosion has ushered in vast numbers of new systems. Each new system presents new challenges to programmers and application developers. Part of the challenge is adapting to new architectures with new performance characteristics. Different vendors release systems with widely varying architectures that perform differently in different situations. Furthermore, since vendors need only provide a single performance number (total MFLOPS, typically for a single benchmark), they only have strong incentive initially to optimize the API of their choice. Consequently, only ... continued below

Physical Description

2127 Kb

Creation Information

Fuller, Douglas James August 1, 2005.

Context

This thesis or dissertation is part of the collection entitled: Office of Scientific & Technical Information Technical Reports and was provided by UNT Libraries Government Documents Department to Digital Library, a digital repository hosted by the UNT Libraries. More information about this document can be viewed below.

Who

People and organizations associated with either the creation of this thesis or dissertation or its content.

Sponsor

Publisher

  • Ames Laboratory
    Publisher Info: AMES (Ames Laboratory (AMES), Ames, IA)
    Place of Publication: Ames, Iowa

Provided By

UNT Libraries Government Documents Department

Serving as both a federal and a state depository library, the UNT Libraries Government Documents Department maintains millions of items in a variety of formats. The department is a member of the FDLP Content Partnerships Program and an Affiliated Archive of the National Archives.

Contact Us

What

Descriptive information to help identify this thesis or dissertation. Follow the links below to find similar items on the Digital Library.

Description

The high performance computing community has experienced an explosive improvement in distributed-shared memory hardware. Driven by increasing real-world problem complexity, this explosion has ushered in vast numbers of new systems. Each new system presents new challenges to programmers and application developers. Part of the challenge is adapting to new architectures with new performance characteristics. Different vendors release systems with widely varying architectures that perform differently in different situations. Furthermore, since vendors need only provide a single performance number (total MFLOPS, typically for a single benchmark), they only have strong incentive initially to optimize the API of their choice. Consequently, only a fraction of the available APIs are well optimized on most systems. This causes issues porting and writing maintainable software, let alone issues for programmers burdened with mastering each new API as it is released. Also, programmers wishing to use a certain machine must choose their API based on the underlying hardware instead of the application. This thesis argues that a flexible, extensible translator for distributed-shared memory APIs can help address some of these issues. For example, a translator might take as input code in one API and output an equivalent program in another. Such a translator could provide instant porting for applications to new systems that do not support the application's library or language natively. While open-source APIs are abundant, they do not perform optimally everywhere. A translator would also allow performance testing using a single base code translated to a number of different APIs. Most significantly, this type of translator frees programmers to select the most appropriate API for a given application based on the application (and developer) itself instead of the underlying hardware.

Physical Description

2127 Kb

Language

Identifier

Unique identifying numbers for this document in the Digital Library or other systems.

  • Report No.: IS-T 2741
  • Grant Number: W-7405-ENG-82
  • Office of Scientific & Technical Information Report Number: 850080
  • Archival Resource Key: ark:/67531/metadc780233

Collections

This document is part of the following collection of related materials.

Office of Scientific & Technical Information Technical Reports

What responsibilities do I have when using this thesis or dissertation?

When

Dates and time periods associated with this thesis or dissertation.

Creation Date

  • August 1, 2005

Added to The UNT Digital Library

  • Dec. 3, 2015, 9:30 a.m.

Description Last Updated

  • Nov. 3, 2016, 11:38 a.m.

Usage Statistics

When was this document last used?

Yesterday: 1
Past 30 days: 1
Total Uses: 3

Interact With This Thesis Or Dissertation

Here are some suggestions for what to do next.

Start Reading

PDF Version Also Available for Download.

Citations, Rights, Re-Use

Fuller, Douglas James. Translation techniques for distributed-shared memory programming models, thesis or dissertation, August 1, 2005; Ames, Iowa. (digital.library.unt.edu/ark:/67531/metadc780233/: accessed August 16, 2017), University of North Texas Libraries, Digital Library, digital.library.unt.edu; crediting UNT Libraries Government Documents Department.