Some theoretical discussions about the unfairness created by different buffer units in routers

Abstract

In recent years, one of the hottest research topics is router buffer sizing. It is very important to size the buffer for routers with very small buffers in future Internet, in particular, in emerging optical networks as optical buffers are needed but very expensive. In 2004, the Stanford research group presented that the core router buffer needs only dozens of packets to meet acceptable link utilization. Then there were a variety of different views presented in the literature. However, all prior studies ignored whether buffer unit is structured in terms of packet or byte. In previous studies, it made no much difference when only TCP traffic was considered. In 2011, our research group studied the buffer sizing in routers with very small buffers when the realtime UDP and TCP traffic coexist. We found that the difference of buffer unit structure obviously impacts the loss performance of routers with very small buffers. When buffer unit is structured in terms of packet, the UDP packet loss rate is higher than the TCP packet loss rate. However, when buffer unit is structured in terms of byte, the UDP packet loss rate is lower than the TCP packet loss rate. We can know the different buffer units would create an unfair impact on UDP traffic. In this paper, we study and analyze this unexpected phenomenon and unfairness, then we present some models independently for some theoretical discussions about the unfairness created by different buffer units in routers.