Premium
Estimating cell loss rates in high‐speed networks with leaky bucket controlled sources
Author(s) -
Johri Pravin K.
Publication year - 1995
Publication title -
international journal of communication systems
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.344
H-Index - 49
eISSN - 1099-1131
pISSN - 1074-5351
DOI - 10.1002/dac.4500080503
Subject(s) - leaky bucket , computer science , buffer overflow , network packet , asymptotic analysis , packet loss , bursting , exponential function , mathematical optimization , real time computing , computer network , mathematics , mathematical analysis , neuroscience , biology
Abstract This paper deals with high‐speed networks, such as the ATM network, over which data is transferred in cells or packets. Results based on asymptotic analysis of stochastic fluid flow models imply that these networks would have to be run at very low utilization to satisfy the stringent performance requirement of very low cell loss rates. Recent papers have quantified that asymptotic approximations can be quite bad with even a modest number of users sending traffic into the network. In this paper it is shown that two more assumptions in the asymptotic analysis may lead to significant overestimation of the cell loss rates. The more significant of these is the use of the underlying exponential ON durations if sources are policed. The second and less significant is the assumption of an infinite buffer. A new approximate nalytic method is then developed for estimating the cell loss rate with sources policed by a leaky bucket type mechanism. This method focuses directly on the problem of measuring cell losses in the presence of many sources simultaneously bursting in a time period small enough to cause buffer overflow. It turns out to be highly accurate even with some rather drastic simplifying assumptions.