Distributed-Memory FastFlow Building Blocks

We present the new distributed-memory run-time system (RTS) of the C++-based open-source structured parallel programming library FastFlow . The new RTS enables the execution of FastFlow shared-memory applications written using its Building Blocks (BBs) on distributed systems with minimal changes to...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:International journal of parallel programming Jg. 51; H. 1; S. 1 - 21
Hauptverfasser: Tonci, Nicolò, Torquati, Massimo, Mencagli, Gabriele, Danelutto, Marco
Format: Journal Article
Sprache:Englisch
Veröffentlicht: New York Springer US 01.02.2023
Springer Nature B.V
Schlagworte:
Algorithms
Code reuse
Communication
Computer networks
Computer Science
Distributed memory
Parallel programming
Processor Architectures
Software
Software Engineering/Programming and Operating Systems
Special Issue on High-Level Parallel Programming and Applications (HLPP 2022)
Theory of Computation
Distributed programming
Algorithmic skeletons
Parallel patterns
High-level parallel programming
Building blocks
ISSN:0885-7458, 1573-7640
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:We present the new distributed-memory run-time system (RTS) of the C++-based open-source structured parallel programming library FastFlow . The new RTS enables the execution of FastFlow shared-memory applications written using its Building Blocks (BBs) on distributed systems with minimal changes to the original program. The changes required are all high-level and deal with introducing distributed groups ( dgroup ), i.e., logical partitions of the BBs composing the application streaming graph. A dgroup , which in turn is implemented using FastFlow ’s BBs, can be deployed and executed on a remote machine and communicate with other dgroups according to the original shared-memory FastFlow streaming programming model. We present how to define the distributed groups and how we faced the problem of data serialization and communication performance tuning through transparent messages’ batching and their scheduling. Finally, we present a study of the overhead introduced by dgroups considering some benchmarks on a sixteen-node cluster.
Bibliographie:ObjectType-Article-1
SourceType-Scholarly Journals-1
ObjectType-Feature-2
content type line 14
ISSN:0885-7458
1573-7640
DOI:10.1007/s10766-022-00750-5