Detecting Interference through Passive Observation
Overview
Interference is the key phenomenon that differentiates wireless
networks for wired networks. The problem of determining the
interference pattern of a wireless network can informally be stated as
follows - Given any subset of links in the network, what is the
performance of each of these links when we try to simultaneously
send
data across all of them? Knowledge of the interference pattern is a
pre-requisite for solving a number of problems such as channel
assignment, scheduling, resource allocation and admission control.
However, most previous work either assumes that the interference
pattern is already available or that it can be approximated using
simple heuristics or models. Recent measurements on testbeds have
shown that such heuristics do not perform very well. While it is indeed
possible to accurately determine the interference pattern
through measurements, such active measurements can be
very expensive both in terms of network resources and time.
In this
work, we present a novel algorithm to avoid this problem by inferring
interference from passively observing the performance of existing
traffic in the network. We have implemented this algorithm in both a
simulator and a testbed and show that it can accurately determine the
interference pattern for active links (that are currently
sending data) within a minute. It works well without modifications for a
wide variety of traffic and network scenarios including different
types of radio channels and antennae.
People
- Ananth
Rajagopala Rao (ananthar <at> cs
<dot> berkeley <dot> edu)
- Ion
Stoica (istoica <at> cs <dot>
berkeley <dot> edu)