Design and implementation of a parallel pipe

  • Authors:

  • Jonathan Karges;Otto Ritter;Sándor Suhai

  • Affiliations:

  • Deutsches Krebsforschungszentrum (DKFZ), Dept. of Molecular Biophysics, Im Neuenheimer Feld 280, D-69120 Heidelberg, Germany;Deutsches Krebsforschungszentrum (DKFZ), Dept. of Molecular Biophysics, Im Neuenheimer Feld 280, D-69120 Heidelberg, Germany;Deutsches Krebsforschungszentrum (DKFZ), Dept. of Molecular Biophysics, Im Neuenheimer Feld 280, D-69120 Heidelberg, Germany

  • Venue:
  • ACM SIGOPS Operating Systems Review
  • Year:
  • 1997

Quantified Score

Hi-index 0.00

Visualization

Abstract

The parallel pipe PAR_PIPE is a communication mechanism. Data which are written to the PAR_PIPE by a single process can be read by multiple processes concurrently in first-in-first-out manner, where all reading processes read the entire data stream. We use a dynamic partitioned buffer limited in size only by available hardware resources to ensure that the writing process cannot be blocked because of slow reading processes. In a further step we examine how far caching with an optimized saving policy on top of the file based buffer can improve the performance of the PAR_PIPE. Design and implementation of two versions of the PAR_PIPE (one without caching, one with caching) are described. We outline the general advantages of the PAR_PIPE over conventional UNIX shell script solutions and evaluate the results of performance measurements done for both PAR_PIPE versions.