CBA: A cluster-based client/server data aggregation routing protocol

Abstract

Client/server routing forwards data samples from the source nodes to the sink through single or multi-hop paths which are formed over a flat or hierarchical infrastructure. Depending on the routing infrastructure, the intermediate nodes may perform in-network data aggregation to collect and combine the data samples which are measures of the environmental events. Minimising energy consumption is a vital requirement due to resource constraints in wireless sensor networks. Data collection delay should be minimised as it is the key to data freshness. At the same time, the number of collected data samples should be maximised, as it should lead to increased accuracy and robustness in data collection. Owing to these, we define the system objective to be maximising the number of delivered data samples, while minimising energy consumption and data collection delay. We propose a cluster-based client/server data aggregation routing protocol called Cluster-Based client/server data Aggregation routing protocol (CBA). It dynamically partitions the network into a set of data-centric clusters using a lightweight clustering approach based on the Hamming distance. The cluster-heads then form a Minimum Spanning Tree (MST) as the network backbone to forward aggregated results to the sink. A parallel collision-guided technique is used to minimise the establishment cost of the tree infrastructure. Compared with the conventional routing protocols like MR-LEACH and Directed DiFFusion, CBA reduces energy consumption and data collection delay and increases accuracy (the number of captured data samples). In addition, our protocol reduces the impacts of the network architecture and the event source distribution model (distributed and centralised) on the performance of data aggregation routing.

Introduction

Wireless Sensor Networks (WSNs) comprise a number of sensor nodes which typically measure and report environmental data. The nodes are typically networked in a self-organising manner without any specific infrastructure or centralised control [54]. The key objective of WSN protocols is to minimise the cost of ambient data collection. Ambient data samples should be collected and forwarded through minimum cost links (in terms of hop count and consumed energy) to data consumer access point (sink) for further analysis and manipulation. The WSN architecture is generally classified in of two ways: distributed (flat) and hierarchical [38]. In the former, the nodes are randomly scattered in the field, whereas in the latter, sensor nodes are organised with a specific distribution topology such as grid, cluster or tree.

The sensor nodes themselves are typically highly resource constrained (in terms of energy, computation, communication and storage) and able to perform three key tasks:[2] (i) measuring a physical quantity (such as temperature or light) from the surrounding environment, (ii) processing (and storing) the sensed data and (iii) transmitting the information to collection points (called Sinks) for either future processing or consumer access.

WSN routing is the field of research that focuses on the interconnection of sensor nodes via either single or multi-hop paths to forward data packets from event regions to the sink. However, the routing overhead increases if raw data packets are forwarded from each source region to the sink. Data aggregation is a technique that collects and combines data packets to express the collected information in a summary form. It reduces the number and size of transmissions and eliminates redundant data packets. WSN Routing can be performed in two ways with data aggregation [47]: mobile agent and client/server. The former routes mobile agent(s) to collect and aggregate data samples from the sensor nodes, whereas the latter establishes an hierarchical network in which data packets are aggregated and forwarded from the ambient event regions to the sink in a convergent manner.

Client/server data aggregation routing establishes the paths according to the network architecture that can be flat or hierarchical [10]. In flat networks, the routes are established from the source nodes converging towards the sink as all nodes play same roles. Apart from sink, intermediate nodes can perform in-network data aggregation if they receive multiple data packets. However, no node is particularly selected to perform data aggregation in flat networks. In hierarchical networks, the nodes play different roles such as network bridge, intermediate aggregator or data consumer access point. Hierarchical routes are usually established from the source nodes to the sink via intermediate nodes which carry out the process of data aggregation.

There are five key issues that need to be considered by designers/developers of WSN client/server data aggregation routing whether flat or hierarchical [39], [52]:

• Energy consumption: power resources need to be used efficiently in WSNs as they are highly constrained. Forwarding data packets over long paths, overhearing and message conflicts/collisions are the behaviours that increase energy consumption in flat WSNs. On the other hand, the cost of establishing and maintaining an hierarchical infrastructure must be minimised if the costs are not to outweigh the benefits.

• Network congestion: simultaneous access to the limited wireless channels increases network congestion and consequently increase the probability of message failures in WSNs. It can increase network resource consumption as the source nodes need to re-transmit failed data packets. Network congestion is decreased in hierarchical networks, as compared to flat, due to the smaller number of nodes which need simultaneously to access the wireless channels. Hierarchical WSNs partition the network into a set of groups in which a small number of nodes (group leaders/representatives) are in charge of managing the group communications. However, network congestion could be problematic in hierarchical WSNs as the number of group and/or leaders increases.

• Overhearing: receiving network packets which do not belong to the receiver nodes increases network resource consumption in WSN. Hierarchical infrastructure has the potential to reduce overhearing (compared to flat networks) as the communications can be locally limited to the node groups. Depending on the size of groups, however, overhearing is increased if groups are large and/or dense.

• Delay: end-to-end delay (ETE) should be minimised in data collection as it is key to data freshness. ETE depends on network traffic and path length (hop count) from the source regions to the sink.

• Data collection/aggregation from Event-Radius (ER) and Random-Source (RS) event sources: The event occurs in a single point of the sensing field in ER (i.e 100% detection), whereas the event sources are randomly distributed in RS (i.e random detection) [22]. RS data collection increases network congestion, delay and resource consumption especially in a flat network, as each source node needs to establish a path to forward data to the sink. It can be resolved in hierarchical networks by grouping the source nodes in which the group representatives forward the aggregated data of group node to the sink. This results in a reduction of routing overhead, network traffic and resource consumption. However, the group leaders miss collecting data samples from source nodes which are not joined to the hierarchical infrastructure. For this reason, the hierarchical infrastructure needs to minimise the establishment and maintenance cost and maximise coverage of event regions either in RS or ER.

In the remainder of this article, Section 2 outlines well-known client/server data aggregation routing protocols to highlight their advantages, features and techniques. Section 3 describes the CBA protocol and the key techniques which are used to enhance the performance and resolve the existing drawbacks of client/server data aggregation routing. Section 4 focuses on the experimental plans to test the performance of CBA. Section 5 evaluates the performance of CBA according to five metrics: total consumed energy, total number of delivered data samples (accuracy), average end-to-end delay, average hop count and total transmitted traffic which are usually used to test the performance of client/server data aggregation routing protocols. The results are measured and discussed to evaluate the performance of CBA in comparison to two client/server data aggregation routing protocols namely MR-LEACH [14](hierarchical) and Directed DiFFusion (DDiFF) [20] (flat). These protocols are selected as they are well-known in the literature, widely simulated and implemented both in the real world and for our chosen experimental platform of OMNET++. This last contributes to the correctness and credibility of our evaluation, because we are able to compare CBA against two client/server routing protocols that have been independently written and verified for OMNET++. Comparison against more recent protocols is also desirable, but this is not feasible without appropriately-verified implementations in OMNET++ yet lacks credibility if authored ourselves. Section 6 concludes the key advantages and disadvantages of CBA protocol and then highlights the research issues which need to be addressed as future works.