A parallelisation tale of two languages

• Tomeu, Antonio J. ^[1] ; Salguero, Alberto G. ^[1] ; Capel, Manuel I. ^[2]
  1. [1] Universidad de Cádiz

     Universidad de Cádiz

     Cádiz, España

     
  2. [2] Universidad de Granada

     Universidad de Granada

     Granada, España

     
• Localización: Annals of Multicore and GPU Programming: AMGP, ISSN 2341-3158, Vol. 2, Nº. 1, 2015, págs. 81-94
• Idioma: inglés
• Enlaces
  • Texto completo (pdf)
• Resumen

  • There are several APIs (C+11, TBB, OpenMPI,... ) that properly support  code parallelism in any of the most used programming languages (C++, C\#,Java, Phyton,  or Erlang) in telecommunications industry nowadays, but it is a complicated issue for a novel programmer of multicores to decide which one to choose in order to get an application parallelized with high performance.C++11 is the new ISO standard for C++, which offers atomic declaration of variables, load/store operators and supports portable multithreaded programming in this familiy of development languages.Java SE has been including new primitives for doing high--level concurrency in Java programs from version 5.0 on. Currently, Java SE 8 also offers parallel syntactical constructs for new multicore architectures programs, such as lambda functions  with an efficient tail recursion implementation or execution to future.Within a similar historical context,  as in the Charles Dickens' A Tale of Two Cities, there may exist equivalent problems in two close--up places but they can be solved in a quite different way. In this respect, we are performing a systematic comparative study between two programming languages that are intensively deployed in industry today, probably due to their popular parallel programming APIs, Java Concurrency Utilities and  C++11 Standard Library, respectively. We carry out a set of measurements to compare the execution time and performance (speedup) of parallel versions of  (a) Belousov-Zhabotinsky chemical reaction simulation with cellular automata and (b) Discrete 2D-Convolution.Differently from other similar work in the field, we focus here more on foretelling performance results that different APIs are offering to non-specialized programmers than in low--level implementation details.