Interference Queuing Networks on Grids
share
Motivated by applications in large scale wireless networks, we introduce and study queuing systems of interacting queues where the interactions mimic wireless interference. Our network model is an abstraction of an ad-hoc wireless network and is intended to capture the interplay between the geometry of wireless links, which determine how the links interfere with each other, and their temporal traffic dynamics. When a link accesses the spectrum, it causes interference to other nearby links, and thus their rate of communication is lowered. This in turn implies they access the spectrum longer and cause interference to nearby links for a larger duration. This coupling between the geometry and temporal dynamics is one of the main challenges in assessing the performance of wireless systems. Most prior work have modeled this phenomenon in a cellular network setting by generalizing the so called coupled processors model. These models are difficult to analyze and even in the simplest case, and one has to typically resort to bounds and approximations to derive insight. We consider the ad-hoc network setting and propose a new model of queues, which interact through interference. We show that this model is tractable, in the sense we can compute certain performance metrics even when there is an infinite number of queues. Specifically, we prove a stability phase-transition for such queuing systems. Furthermore, we also derive exactly the mean number of customers in any queue in steady state. Thanks to Little's Law, this also yields an exact expression for mean delay of a typical link in steady state. The key to our results is the analysis of certain Rate Conservation equations along with monotonicity and tightness arguments, which are new and interesting in their own right. Our model and results thus provide a basis to evaluate the performance of various protocols on large networks.
READ FULL TEXT