skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Some language issues in high performance computing: translation from fine-grained parallelism to coarse-grained parallelism.

Authors:
; ;
Publication Date:
Research Org.:
Sandia National Laboratories
Sponsoring Org.:
USDOE
OSTI Identifier:
882922
Report Number(s):
SAND2006-0422
DOE Contract Number:
AC04-94AL85000
Resource Type:
Technical Report
Country of Publication:
United States
Language:
English
Subject:
Parallel computers-Programming.; Parallel programming.; High performance computing.

Citation Formats

Goudy, Susan Phelps, Wen, Zhaofang., and Huang, Shan Shan. Some language issues in high performance computing: translation from fine-grained parallelism to coarse-grained parallelism.. United States: N. p., 2006. Web. doi:10.2172/882922.
Goudy, Susan Phelps, Wen, Zhaofang., & Huang, Shan Shan. Some language issues in high performance computing: translation from fine-grained parallelism to coarse-grained parallelism.. United States. doi:10.2172/882922.
Goudy, Susan Phelps, Wen, Zhaofang., and Huang, Shan Shan. Sun . "Some language issues in high performance computing: translation from fine-grained parallelism to coarse-grained parallelism.". United States. doi:10.2172/882922. https://www.osti.gov/servlets/purl/882922.
@article{osti_882922,
title = {Some language issues in high performance computing: translation from fine-grained parallelism to coarse-grained parallelism.},
author = {Goudy, Susan Phelps and Wen, Zhaofang. and Huang, Shan Shan},
abstractNote = {},
doi = {10.2172/882922},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Sun Jan 01 00:00:00 EST 2006},
month = {Sun Jan 01 00:00:00 EST 2006}
}

Technical Report:

Save / Share:
  • Abstract not provided.
  • Machine teaming methods are becoming accepted as additions to the biologist`s data-analysis tool kit. However, scaling these techniques up to large data sets, such as those in biological and medical domains, is problematic in terms of both the required computational search effort and required memory (and the detrimental effects of excessive swapping). Our approach to tackling the problem of scaling up to large datasets is to take advantage of the ubiquitous workstation networks that are generally available in scientific and engineering environments. This paper introduces the notion of the invariant-partitioning property--that for certain evaluation criteria it is possible to partitionmore » a data set across multiple processors such that any rule that is satisfactory over the entire data set will also be satisfactory on at least one subset. In addition, by taking advantage of cooperation through interprocess communication, it is possible to build distributed learning algorithms such that only rules that are satisfactory over the entire data set will be learned. We describe a distributed learning system, CorPRL, that takes advantage of the invariant-partitioning property to learn from very large data sets, and present results demonstrating CorPRL`s effectiveness in analyzing data from two databases.« less
  • The work explored the linking of modern developing machine learning techniques (manifold learning and in particular diffusion maps) with traditional PDE modeling/discretization/scientific computation techniques via the equation-free methodology developed by the PI. The result (in addition to several PhD degrees, two of them by CSGF Fellows) was a sequence of strong developments - in part on the algorithmic side, linking data mining with scientific computing, and in part on applications, ranging from PDE discretizations to molecular dynamics and complex network dynamics.
  • In the search for ''good'' parallel programming environments for Sandia's current and future parallel architectures, they revisit a long-standing open question. Can the PRAM parallel algorithms designed by theoretical computer scientists over the last two decades be implemented efficiently? This open question has co-existed with ongoing efforts in the HPC community to develop practical parallel programming models that can simultaneously provide ease of use, expressiveness, performance, and scalability. Unfortunately, no single model has met all these competing requirements. Here they propose a parallel programming environment, PRAM C, to bridge the gap between theory and practice. This is an attempt tomore » provide an affirmative answer to the PRAM question, and to satisfy these competing practical requirements. This environment consists of a new thin runtime layer and an ANSI C extension. The C extension has two control constructs and one additional data type concept, ''shared''. This C extension should enable easy translation from PRAM algorithms to real parallel programs, much like the translation from sequential algorithms to C programs. The thin runtime layer bundles fine-grained communication requests into coarse-grained communication to be served by message-passing. Although the PRAM represents SIMD-style fine-grained parallelism, a stand-alone PRAM C environment can support both fine-grained and coarse-grained parallel programming in either a MIMD or SPMD style, interoperate with existing MPI libraries, and use existing hardware. The PRAM C model can also be integrated easily with existing models. Unlike related efforts proposing innovative hardware with the goal to realize the PRAM, ours can be a pure software solution with the purpose to provide a practical programming environment for existing parallel machines; it also has the potential to perform well on future parallel architectures.« less