Reduction in Packet Delay Through the use of Common Buffer over Distributed Buffer in the Routing Node of NOC Architecture

TL;DR

This paper evaluates the impact of using a common buffer instead of distributed buffers in NOC routing nodes, demonstrating significant latency reductions through modeling, simulation, and hardware synthesis.

Contribution

It introduces a comparative analysis of common versus distributed buffers in NOC routers, quantifies latency improvements, and provides models and hardware implementation details.

Findings

Common buffer reduces latency by approximately 40-46%.

Simulation and modeling confirm latency improvements with common buffer.

Hardware synthesis validates practical feasibility of the proposed buffer design.

Abstract

Performance evaluation of the routing node in terms of latency is the characteristics of an efficient design of Buffer in input module. It is intended to study and quantify the behavior of the single packet array design in relation to the multiple packet array design. The utilization efficiency of the packet buffer array improves when a common buffer is used instead of individual buffers in each input port. First Poissons Queuing model was prepared to manifest the differences in packet delays. The queuing model can be classified as (M/M/1), (32/FIFO). Arrival rate has been assumed to be Poisson distributed with a mean arrival rate of 10 x 1000000. The service rate is assumed to be exponentially distributed with a mean service rate of 10.05 x 1000000. It has been observed that latency in Common Buffer improved by 46 percent over its distributed buffer. A Simulink model later simulated on MATLAB to calculate the improvement in packet delay. It has been observed that the delay improved by approximately 40 percent through the use of a common buffer. A verilog RTL for both common and shared buffer has been prepared and later synthesized using Design Compiler of SYNOPSYS. In distributed buffer, arrival of data packet could be delayed by 2 or 4 clock cycles which lead to latency improvement either by 17 percent or 34 percent in a common buffer