Abstract
It is now accepted that parallelism will be the primary means of increasing performance of computer systems, but the preoccupation over the last 30 years with frequency scaling of single-core systems has meant that relatively little work has taken place with regards to how we should architect and program parallel computers. Currently, there is little consensus in the approach taken with new designs, which has resulted in a growing gap between the design of languages and architectures, and the algorithms written. This paper argues that to be able to effectively use parallel systems, simple concurrency mechanisms must be provided
as features of a programming language, and that these must be supported by primitive operations in the underlying architecture. In particular, the concepts of process migration and parallel recursion allow the expression of simple but powerful concurrent programs.
Translated title of the contribution | Dynamic generation of parallel computations |
---|---|
Original language | English |
Title of host publication | The 2010 UK Electronics Forum |
Editors | A. Kolemans, A. Yakovlev |
Publisher | Newcastle University |
Pages | 7 - 17 |
Number of pages | 10 |
ISBN (Print) | 9780701702328 |
Publication status | Published - Jun 2010 |
Bibliographical note
Name and Venue of Event: UK Electronics ForumMedium/genre: Workshop Paper