Time-division multiplexing vs network calculus: A comparison

Wolfgang Puffitsch, Rasmus Bo Sørensen, Martin Schoeberl

Research output: Chapter in Book/Report/Conference proceedingArticle in proceedingsResearchpeer-review

Abstract

Networks-on-chip are increasingly common in modern multicore architectures. However, general-purpose networks-on-chip are not always well suited for real-time applications that require bandwidth and latency guarantees. Two approaches to provide real-time guarantees have emerged: time-division multiplexing, where traffic is scheduled according to a precalculated static schedule, and network calculus, a mathematical framework to reason about dynamically scheduled networks. This paper compares the two approaches to provide insight into their relative advantages and disadvantages. The results show that time-division multiplexing leads to better worst-case latencies, while network calculus supports higher bandwidths. Furthermore, time-division multiplexing leads to a simpler hardware implementation, while dynamically scheduled networks-on-chip allow the integration of best-effort traffic in the on-chip network in a more natural way.
Original languageEnglish
Title of host publicationProceedings of the 23rd International Conference on Real Time and Networks Systems (RTNS 2015)
PublisherAssociation for Computing Machinery
Publication date2015
Pages289-296
ISBN (Print)978-1-4503-3591-1
DOIs
Publication statusPublished - 2015
Event23rd International Conference on Real-Time Networks and Systems - Lille, France
Duration: 4 Nov 20156 Nov 2015
Conference number: 23

Conference

Conference23rd International Conference on Real-Time Networks and Systems
Number23
CountryFrance
CityLille
Period04/11/201506/11/2015

Keywords

  • Networks
  • Network on chip
  • Network performance modeling
  • Computer systems organization
  • Real-time systems

Cite this

Puffitsch, W., Sørensen, R. B., & Schoeberl, M. (2015). Time-division multiplexing vs network calculus: A comparison. In Proceedings of the 23rd International Conference on Real Time and Networks Systems (RTNS 2015) (pp. 289-296). Association for Computing Machinery. https://doi.org/10.1145/2834848.2834868