2nd International ICST Conference on Simulation Tools and Techniques

Research Article

Comparison of Network-on-Chip Topologies for Multicore Systems Considering Multicast and Local Traffic

Download804 downloads
  • @INPROCEEDINGS{10.4108/ICST.SIMUTOOLS2009.5590,
        author={Dietmar Tutsch and Miroslaw Malek},
        title={Comparison of Network-on-Chip Topologies for Multicore Systems Considering Multicast and Local Traffic},
        proceedings={2nd International ICST Conference on Simulation Tools and Techniques},
        publisher={ICST},
        proceedings_a={SIMUTOOLS},
        year={2010},
        month={5},
        keywords={network-on-chip multicore processor multicast simulation performance},
        doi={10.4108/ICST.SIMUTOOLS2009.5590}
    }
    
  • Dietmar Tutsch
    Miroslaw Malek
    Year: 2010
    Comparison of Network-on-Chip Topologies for Multicore Systems Considering Multicast and Local Traffic
    SIMUTOOLS
    ICST
    DOI: 10.4108/ICST.SIMUTOOLS2009.5590
Dietmar Tutsch1,*, Miroslaw Malek2,*
  • 1: Bergische Universität Wuppertal, 42119 Wuppertal, Germany
  • 2: Humboldt-Universität zu Berlin, 12489 Berlin, Germany
*Contact email: tutsch@uni-wuppertal.de, malek@informatik.hu-berlin.de

Abstract

Performance of two network-on-chip (NoC) topologies is compared for the use in multicore processors. The performance evaluation is supported by the CINSim simulator. This simulator has been developed to model a variety of network topologies that are based on atomic components such as buffers, routers, traffic generators, and target buffers. The development of this simulator was driven by the investigation of networks-on-chip. But off-chip networks can be examined as well. Two examples for NoC topologies, a mesh and a bidirectional interconnection network, are compared. Unicast traffic is used as well as multicast and local traffic, which both represent a significant part of the network traffic for evaluating multicore processors. In addition to the performance, the mean distance, the diameter, and the buffer cost are calculated for both network topologies. The results show that bidirectional multistage interconnection networks outperform meshes. A clearly better scalability is shown by the bidirectional multistage interconnection networks.