The Multicore Software Challenge

  • Tagung:

    Vortrag 

  • Autoren:

    Walter F. Tichy 

  • Summary

    With multicore processors, servers, PCs, and laptops have become truly parallel machines. The mobile phone and embedded applications will follow. Hardware manufacturers predict a hundred processors and more per chip. General-purpose, parallel software, on the other hand, is scarce. What to do with all the processors?This briefing will overview the current hardware developments, provide examples of successfully parallelized, non-numeric applications, and discuss the lessons learnt for software engineering. The good news is that parallelization is not a black art—it can be handled with reasonable effort. However, significant restructurings are typically required when parallelizing existing, serial applications. We also present recent advances in the areas of automatic performance tuning and programming languages that allow the succinct expression of frequent parallel patterns. A comparative case study will shed light on the actual advantages of transactional memory over locking. An outlook on future tools and methods for parallelization will be given.Time permitting, a summary of the International Workshop of Multicore Software Engineering will be presented. 

  • Jahr:

    2009 

Beteiligte Mitarbeiter (zufällige Reihenfolge)
Titel Vorname Nachname

Bibtex

@vortrag{,
author={Walter F. Tichy},
title={The Multicore Software Challenge},
year=2009,
month=May,
abstract={With multicore processors, servers, PCs, and laptops have become truly parallel machines. The mobile phone and embedded applications will follow. Hardware manufacturers predict a hundred processors and more per chip. General-purpose, parallel software, on the other hand, is scarce. What to do with all the processors?This briefing will overview the current hardware developments, provide examples of successfully parallelized, non-numeric applications, and discuss the lessons learnt for software engineering. The good news is that parallelization is not a black art—it can be handled with reasonable effort. However, significant restructurings are typically required when parallelizing existing, serial applications. We also present recent advances in the areas of automatic performance tuning and programming languages that allow the succinct expression of frequent parallel patterns. A comparative case study will shed light on the actual advantages of transactional memory over locking. An outlook on future tools and methods for parallelization will be given.Time permitting, a summary of the International Workshop of Multicore Software Engineering will be presented.},
organization={International Conference on Software Engineering},
address={Vancouver, Canada},