Communication modeling of multicast in all-port wormhole-routed NoCs

Abstract

Multicast is one of the most frequently used collective communication operations in multi-core SoC platforms. Bus as the traditional interconnect architecture for SoC development has been highly efficient in delivering multicast messages. Since the bus is non-scalable, it can not address the bandwidth requirements of the large SoCs. The networks on-chip (NoCs) emerged as a scalable alternative to address the increasing communication demands of such systems. However, due to its hop-to-hop communication, the NoCs may not be able to deliver multicast operations as efficiently as buses do. Adopting multi-port routers has been an approach to improve the performance of the multicast operations in interconnection networks. This paper presents a novel analytical model to compute communication latency of the multicast operation in wormhole-routed interconnection networks employing asynchronous multi-port routers scheme. The model is applied to the Quarc NoC and its validity is verified by comparing the model predictions against the results obtained from a discrete-event simulator developed using OMNET++.

Introduction

Traditionally, the dominant interconnection architectures for systems-on-chip (SoC) has been bus-based. Driven by an exponential downscaling of feature size (following Moore’s law), SoCs consisting of billions of gates and hundreds of processing units are becoming a reality. Since the bus is inherently non-scalable and ad-hoc interconnection of the IP cores is not affordable in large SoCs, adopting other alternatives to serve as communication medium for SoC development has become inevitable. Networks on-Chip (NoCs) have been emerged as the most promising solution (Dally et al., 2001, Benini and De Micheli, 2002). NoCs offer a scalable (Guerriert, 2001, Rijpkema et al., 2003), structured (Benini and De Micheli, 2002), power efficient and reliable (Benini and De Micheli, 2001, Bolotin et al., 2004) communication medium to address the technological and design issues regarding to development of large SoCs.

The bus-based SoC development has been able to leverage the advantages of a shared medium to efficiently implement most operations of a collective nature. An example of such operations is cache coherency, which is based on multicasting a message to all or a number of processing cores.

Unlike the bus-based architectures, the communication in an NoC-based platform is hop-by-hop in nature and efficient implementation of collective communication operations is still a challenge. NoCs are in principle similar to interconnection networks for parallel computers with multiple processors. Therefore, the NoC community can leverage the results of a large body of research available in that domain to address efficient implementation of the collective communication in the NoC realm.

Analytical modeling has been widely used to evaluate performance of the communication in interconnection networks. The literature has witnessed numerous analytical performance models of unicast traffic (Draper and Ghosh, 1994, Moadeli et al., 2007) and analysis of unicast traffic in presence of broadcast traffic (Shahrabi et al., 2003) in parallel computers and NoC domains. (Shahrabi et al., 2000) introduced a model for computing broadcast communication latency in a Hypercube-topology network. However, in their system model only unicast traffic was wormhole-routed, and broadcast communication was not wormhole-routed. Also, their model was developed for architectures using one-port routers. Moadeli et al. (2009) presented a model to compute average message latency of the multicast communication in NoCs using all-port routers. This work constitutes an improvement to the earlier work by presenting a more detailed representation of the analytical model, adding the analysis of broadcast along with the multicast communication, and also presenting a more detailed description of the traffic in the Quarc NoC.

The paper is organized as follows. The next section presents the related work to offer multicast. Section 3 introduces a method for analyzing the average message latency of multicast communication in all-port wormhole-routed interconnection networks. Section 4 presents a brief description of the Quarc NoC and the broadcast/multicast routing algorithm in the architecture. Section 5 compares our analytical evaluation with simulation results and finally, in Section 6 the conclusion and future works are presented.