cWRR: A Scheduling Algorithm for Maximizing the Performance of Quality of Service Network Router

Abstract

This paper presents a novel approach for implementing quality of service as demanded by evolving applications in the Internet. For some decades now, research efforts have led to the extension of the TCP/IP in order to make the Internet a full-fledged quality of service network. Novel in the extension is the invention of the Integrated services and Differentiated services architectures. The Differentiated services architecture was widely accepted among researchers because of its scalability. In order to achieve some of the refinements to the current TCP/IP protocol by the IETF for DiffServ implementation in the Internet, new traffic management mechanisms such as differential packet buffering cum differential allocation of available link bandwidth are needed. This report studied some suggested scheduling algorithms in literature on how to incorporate a multi-queue paradigm and enforce service level agreement in the Internet. A new scheduling model that ensures maximum utilization of network bandwidth is used to assess experimental implementation of Differentiated services in a QoS-based router. The model, termed, carry-on Weighted Round Robin (cWRR) proved better than the original Weighted Round Robin (WRR) scheme in terms of low higher throughput and fairness to traffic sources in a multi-queue network core router paradigm. General TermsDifferentiated Services, Quality of Services, and Weighed Round Robin KeywordsDiffServ, QoS Router, Scheduler, TCP/IP, WRR 1.INTRODUCTIONA Router is an indispensable network device that receives traffic streams coming from various sources or other routers and forwards them to the next intermediate router or destination router. A key function of this device operating under a multi-queue paradigm is the scheduling approach through which it decides which packets and from which queue to transmit next. According to [1], the next generation Internet router can implement up to 64 different queues, each queue keeps packets of a particular group of application that expects same treatment from the network. In order to grant certain required quality to a network application, service level agreements between the traffic sources and the network needs to be enforced at the routers. Accordingly, the source specifies a particular request (traffic spec) while the network determines the estimated the sending profile of the source.Any violation may lead to unexpected treatment of the extra traffics from the source. The authors in [2] places emphasis on the importance of a router scheduling algorithm to be a procedure that influences three orthogonal traffic management functions: buffering of packets, ordering of packets for transmission and controlling of congestion by dropping packets if need be. These key functions of a network router scheduling algorithm have led researchers to propose different algorithms for supporting many queues and satisfying various requirements of each traffic group in DiffServ. Most of the algorithms are criticized in terms of flow isolation, delays experienced by...