A.J. Ganesh, P.B. Key, D. Polis, and R. Srikant
There has been much interest in admission control schemes that place the burden of admission control decisions on the end users. In these schemes, referred to as Endpoint Admission Control, the decision to join the network is taken by the user, based on the probing of the network using probe packets. Depending on the level of congestion, routers mark the probe packets and thus inform the user of the state of the network. In this paper, we analyze three mechanisms for providing Endpoint Admission Control: virtual-queue marking, random-early marking and tail drop. For each scheme, we analyze the probing duration necessary to guarantee the required QoS and achieve high link utilization. Our main conclusion is that very few probe packets have to be sent when early marking is used, whereas tail drop requires a large number of probe packets.
In IEEE/ACM Transactions on Networking