A Distributed Feedback Control Mechanism for Quality-of-Service Maintenance in Wireless Sensor Networks

Abstract

Wireless sensor networks are typically operating in a dynamic context where events, such as moving sensor nodes and changing external interference, constantly impact the qualityof-service of the network. We present a distributed feedback control mechanism that actively balances multiple conflicting network-wide quality metrics, such as power consumption and end-to-end packet latency, for a heterogeneous wireless sensor network operating in a dynamic context. Nodes constantly decide if and how to adapt controllable parameters of the entire protocol stack, using sufficient information of the current network state. Using experiments with an actual deployment we show that our controller allows to maintain the required network-wide qualityof-service, with up to 30% less power consumed, compared to the most applicable (re-)configuration approaches. I. MOTIVATIONControllable parameters of individual sensor nodes in a Wireless Sensor Network (WSN), such as the radio transmission power, MAC protocol duty cycle and selected routing parent(s), determine the behaviour of the network. The values of these parameters, referred to as the configuration, determine to what extent expectations on quality metrics, i.e., the required Quality-of-Service (QoS), are satisfied. WSNs typically operate in a dynamic environment or exhibit dynamic behaviour themselves causing the configuration required to achieve sufficient QoS to vary over time. Statically configured nodes [5], [6] cannot cope with dynamics; run-time adaptation of parameters, or reconfiguration, is needed. Current approaches focus either on a single metric [4], [16], typically energy consumption, or on optimizing local metrics [13], [18] and often consider a homogenous network. In practice, the requirements of the applications are typically expressed by multiple conflicting quality metrics expressed on a networkwide scale [3], [14], such as a maximum end-to-end latency of 100 milliseconds, a packet delivery ratio of 90% or maximal network lifetime. There is only limited existing work on runtime adaptation techniques that consider multiple QoS metrics expressed on a network-wide scale. The recent work of [19] proposes a centralized approach which determines how to adapt MAC protocol parameters based on centrally collected network QoS information. In practice, centralized approaches might be too costly, non-scalable or not able to respond quick enough to dynamic events. By focusing on only the MAC protocol, it does not exploit the fact that the parameters from all protocols may influence the behaviour of the network [7].As a homogenous configuration is assumed where all nodes use the same MAC parameters, it furthermore ignores the typical heterogeneity in WSNs and resulting variation of the impact of nodes on the trade-offs involved.In this extended abstract, we introduce a distributed feedback control mechanism to maintain a required QoS, defined by multiple quality metrics, for a WSN in a dynamic context. The approach separates the adaptivity from protocol operation and lets...

A Distributed Reconfiguration Approach for Quality-of-Service Provisioning in Dynamic Heterogeneous Wireless Sensor Networks

“…Parameters are quickly adapted completely until one end of the suitable range to reduce the loss, because the impact of a single adaptation step is not fully observed before the next adaptation. Due to the use of a suitable range in the predictive model, we do not observe both a repeated under and overshooting of the target QoS, as was observed with using a static model which allows the parameter to take every possible value from its range [Steine et al 2012]. The dynamic model does not have the opportunity to adapt (as the time between adaptations is less than the averaging time of 60 seconds), and the model used for the old situation is constantly used.…”

“…Currently, run-time reconfiguration approaches typically rely on simplistic static impact models, where every adaptation step of a parameter is assumed to have the same impact on the metrics [Steine et al 2012]. For such a static model, only a single value needs to be stored for every parameter-metric pair and no run-time overhead is involved with maintaining this model.…”

“…An earlier version of the reconfiguration approach as discussed in this paper was published in [Steine et al 2012]. This paper incorporates and extends [Steine et al 2012], with the following new contributions: -A more elaborated discussion on how the distributed service as introduced in [Steine et al 2011a] allows every node to estimate the network QoS information needed.…”

“…This paper incorporates and extends [Steine et al 2012], with the following new contributions: -A more elaborated discussion on how the distributed service as introduced in [Steine et al 2011a] allows every node to estimate the network QoS information needed. -Instead of optimizing node power consumption we optimize the network lifetime, since minimizing power consumption is not necessarily the same as maximizing the network lifetime.…”

“…-Instead of the simplistic static linear models used in [Steine et al 2012], new adaptive predictive models are used to determine how parameters should be adapted to achieve a desired change in QoS. A feedback control approach is integrated, within the already used feedback control strategy to control the configuration, to maintain accurate model parameters based on the current QoS and QoS before reconfiguration.…”