Interpreting parallel processor performance measurements
SIAM Journal on Scientific and Statistical Computing
Communications of the ACM
Performance measurement of parallel Ada: an applications based approach
Proceedings of the working group on Ada performance issues 1990
Panel: Ada and high performance computing—reality or myth?
WADAS '92 Proceedings of the ninth Washington Ada symposium on Ada: Empowering software users and developers
Multi-core + multi-tasking = multi-opportunity?
Proceedings of the 2007 ACM international conference on SIGAda annual international conference
Hi-index | 0.00 |
This paper reports on experimental results which demonstrate the potential of Ada as a parallel programming language for large scale, scientific applications on high performance multiprocessors. Reported performance results show a linear speed-up of a factor of 10 over 10 processors. A linear speed-up performance over a larger number of processors is indicated given the availability of higher performance configurations and larger data sets.The results were obtained for an Ada program written to take advantage of the multiprocessing of the Sequent Balance and Symmetry computer series, which performed the classification of stars in a large infrared astronomical database, (IRAS). The classification procedure and algorithms, which were developed at NASA Ames Research Center and were originally coded in a Lisp program called AutoClassII, use an approximate Bayesian criterion. The procedure has been applied to a wide variety of classification domains besides astronomy.Through its tasking structure, Ada possesses the intrinsic ability to support parallel processing at the language level; however, much remains to be learned about the implementation of concurrent software systems in Ada on multiprocessor architectures. The paper provides a case study of the design issues faced in the attempt to exploit the inherent concurrency of the procedure and algorithms in the implementation of a portable version of AutoCIassII in Ada for benchmarking multiprocessors.