A QoS-enhanced smart packet scheduler for multi-core processors in intelligent routers using machine learning

Abstract

In this paper, we have developed a QoS-enhanced smart scheduler for multi-core processor in intelligent IP routers using machine (deep) learning. The proposed packet scheduler is stochastic in nature, it can process real-time traffic packets, and it is reconfigurable one. Each core maintains a guaranteed steady-state core load distribution ratio with an aim of keeping the processor utilization close to 100%. The scheduler has the advantages of having each core processing the incoming traffic in a utilization-driven deadline-aware mode, and the load imbalance is minimized dynamically at run-time in an intelligent way using our proposed machine (deep) learning algorithm in order to retain a steady-state load distribution among the cores. This results in minimization of computational overhead to each core, higher throughput, lower value of mean waiting time, and PLR.