Access network buffer dimensioning for bulk TCP traffic

It is largely recognised in the research community that access and edge network buffers need special attention with respect to dimensioning, since those are often attached to bottleneck links in contrast with the core Internet network. The quality of experience of the end user – as a result of the performance of its applications – will depend on the adequate construction of these buffers. The capacity of the access networks keeps increasing. However, there is recently a well-defined share in the network traffic mix of more and more (audio and video) streamed applications. Also on top of traditional web applications, the bandwidth-greedy peer-to-peer elastic (TCP-controlled) traffic is growing. In this paper we consider a best-effort queue with TCP traffic, through which no stringent quality guarantee can be provided to streamed multimedia (non-responsive and time-critical) applications; hence a different buffer is recommended for the real-time applications but this is subject to a different study (e.g., [27]). Given that the last-mile connection is not highspeed, the maximised utilisation of its (download) capacity is an important issue, as well as that each flow gets its fair share. Long-lived, also called “persistent” or bulk traffic flows, create the heaviest buffer demand (and not the short-lived (web) flows). We develop a heuristic formula to calculate the required buffer size, so that both requirements can be fulfilled by choosing the buffer large enough so that there is no packet loss incurred by it. Unfortunately, this buffer size may turn out to be too large to be implemented in state-of-the-art access equipment. So, we also show how efficiency and fairness requirements are met in a smaller buffer in case the buffer acceptance discipline is Drop-tail or RED. We show that Drop-tail cannot achieve fairness and that RED, if appropriately dimensioned, can (and we also present a methodology for its configuration)