Performance of back-off algorithms

Abstract

A well known defect of the standard exponential back-off algorithm used in IEEE 802.11 is its short-term unfairness between users. A user will have high throughput shortly after a successful transmitting but the user will have a low throughput temporarily after backing off again and again because of multiple collisions. This variation of short-term throughput induces unacceptable delay variation for the users that have real-time service demands such as Audio or Video. Recently algorithms like Idle Sense [1] has been proposed to solve this problem by reducing the short-term unfairness between users. It lets each user adapt its probability p of accessing the channel to a predetermined optimal p* based on its observation of traffic on the channel. In this thesis, we discuss some improvements we made to Idle Sense by including a deterministic back-off of [(1 -- p)/p] time slots after a successful transmission. Our algorithm may be viewed as a combination of Idle Sense and Zero Collision [2]. Simulation results show the throughput our algorithm obtains is almost the same as that of Idle Sense but short-term unfairness is reduced. Packets of each user can be transmitted at much more regular intervals and in simulations for Voice over IP (VOIP), the number of VOIP users with a good Quality of Service (QoS) through a single access point can be substantially increased.