A Packetized Round Robin Service Discipline for IP Audio Streams in MPLS DiffServ Networks

Real-time voice applications generate network traffic streams that require stringent perconnection performance guarantees in terms of throughput, delay, delay-jitter, and packet lossrate. To guarantee such performance requirements two levels of control are needed: connection admission control (CAC) at the connection level and service discipline at the packet level. These two levels of control can now be achieved using Multi Protocol Label Switching (MPLS) in conjunction with differentiated services (DiffServ) architecture. As more and more real time voice applications are made available in the Internet, the problem of providing QoS guarantees to such applications becomes more relevant. One difficulty with the Internet is the variable packet sizes allowed, making it harder to calculate delay bounds on various traffic flows. In this paper, the authors present a packetized version of round robin (RR) scheduling discipline to support transmission of voice traffic over an MPLS DiffServ network. It extends previous work by the authors that used a form of Round Robin service discipline for ATM networks. The scheme was called Cell by Cell Round Robin (CRR), and it exploited the fact that ATM cells are of fixed size, providing delay bounds for various streams, fairness and isolation between the streams. We consider now packets of various sizes, mainly due to different encoding schemes. However, we show that, despite the variable sizes of packets, the guarantees provided by the CRR scheme are still valid if traffic classes are defined carefully and some discipline imposed on voice CODECs.