Citation Context

...y and bandwidth bounds on end-to-end traffic flows transported by a best-effort wormhole NoC fabric with no special hardware support for RT traffic. Our approach is inspired by the work by Lee et al. =-=[10]-=- for traditional multiprocessor networks, which we extend in several directions. We propose two different methods for characterizing worst-case packet injection. The first method, RTB-HB (Real-Time Bo...

Citation Context

...traffic conditions. 1. Introduction Wormhole switching [7] has become the dominating switching technique used in contemporary multicomputers and more recently in clusters [8] and system area networks =-=[14]-=-. This is because it requires minimum buffer space and it makes message latency almost independent of the message distance in absence of blocking. In wormhole switching, a message is broken into flits...

Citation Context

... wormhole switching 1. INTRODUCTION Wormhole switching [7] has become the dominating switching technique used in contemporary multicomputers and more recently in clusters [8] and system area networks =-=[14]-=-. This is because it requires minimum buffer space and it makes message latency almost independent of the message distance in absence of blocking. In wormhole switching, a message is broken into flits...

Citation Context

...such streams. This helps in choosing topologies that meet the RT constraints with minimum area, power overhead and optimum utilization of resources. Our approach is inspired by the work by Lee et al. =-=[10]-=- for traditional multiprocessor networks, and extends it in several directions. We propose two different methods for characterizing worst case performance. The first method, Real-Time Bound for High-B...

Citation Context

...e important than seeking high throughput. Yue [24] proposed a Hungarian-based approach to minimize the application’s timing behavior change from its initial deployment after reconfiguration with the assumption that there is no network congestion on the NoC. However, the current and most widely used NoC platform only supports best-effort service [18, 19]. In other words, packages often compete for routing and transmission on a link and congestion will occur if two packages compete for the same link simultaneously. If congestion happens, the package transmission time delay becomes unpredictable [15]. To track the traffic dynamics in NoC platform, Bogdan [3] developed a statistical physics inspired model to capture the non-stationary effects in NoC traffic. As real-time embedded applications require deadline guarantees, researchers have started to address the communication congestion issue by adding advanced features in NoC architectures to support guaranteed levels of service. For instance, Goossens [11] developed the Æthereal NoC using pipelined time-division-multiplexed circuit switching technique which provides guaranteed service by using time slot reservation. Bjerregaard [2] develop...

Citation Context

...d to be adapted for networks on chip due to restricted buffer space in networks on chip. There are also several notable other approaches for quality of service in packet-switched networks [22], [24], =-=[14]-=-, [1], [15]. Qian et al. in [21] use network calculus to derive worst-case delay bounds of best-effort packets, however, the delay bounds could be conservative when several flows intersect each other....

Citation Context

...d to be adapted for networks on chip due to restricted buffer space in networks on chip. There are also several notable other approaches for quality of service in packet-switched networks [22], [24], =-=[14]-=-, [1], [15]. Qian et al. in [21] use network calculus to derive worst-case delay bounds of best-effort packets, however, the delay bounds could be conservative when several flows intersect each other....