Anthony Fynn, [email protected] emergence of wireless sensor networks in the past 15 years can be characterized by the amount ofresearch papers published relating to the field. The research papers also characterize the difficulty faced byresearchers in selecting appropriate techniques to evaluate their work. In this paper, evaluation techniquespresented in research papers on the most important areas of wireless sensor networks are reviewed for theirdeficiencies as well as their effectiveness.Keywords: Wireless Sensor Networks, Archival Storage, Empirical data, Ad-hoc Network, Link Protocols,Network Protocols, Shortest Path, Minimum Transmission1 Introduction2 Storage2.1 Experimental Set-up2.2 Results2.3 Critique2.3.1 Data2.3.2 Workloads2.3.3 Factors2.3.4 Metrics3 Routing3.1 Experimental Set-up3.2 Results3.3 Critique3.3.1 Data3.3.2 Workloads3.3.3 Factors3.3.4 Metrics4 Real-Time Communication4.1 Experimental Set-up4.2 Results4.3 Critique4.3.1 Data4.3.2 Workloads4.3.3 Factors4.3.4 Metrics5 Power Management5.1 Experimental Set-up5.2 Results5.3 CritiquePerformance Analysis of Wireless Sensor Networkshttp://www.cse.wustl.edu/~jain/cse567-06/ftp/sensor_perf/index.html 1 of 125.3.1 Data5.3.2 Workloads5.3.3 Factors5.3.4 Metrics6 Architecture6.1 Experimental Set-up6.2 Results6.3 Critique6.3.1 Data6.3.2 Workloads6.3.3 Factors6.3.4 Metrics7 SummaryReferencesList of AcronymsWireless sensor networks (WSN) have emerged as one of the most exciting fields in Computer Scienceresearch over the past 15 years. Processors with on-board sensors are said to be nearing the size of a dust.Applications of WSN include military surveillance, habitat monitoring, structural monitoring and cargotracking. The evolution of the field may be followed in research papers written by prominent personalities andinstitutions in Computer Science research. A macrocosm of topics in those papers is surveyed and theirevaluation techniques are assessed in this paper. The topics include storage, routing, real-time communication,power management and architecture.These topics are discussed in the following sections. The discussion will be organized in five sections and eachsection will be focused on a research paper that presents an implementation relating to the topic. In eachsection, a general introduction to the topic is given. The introduction is followed by a summary of theevaluation of the implementation as presented in the research paper. This will span two subsections-experimental set-up and results. The following subsection, critique, reviews the evaluation techniques underfour criteria: data, workloads, factors and metrics selected.Back to Table of ContentsThe emergence of many kinds of networked data-centric sensor applications has given more importance todata generated by the sensors. Sensors in these applications probe the environment for useful data for analysis.To achieve a useful infrastructure for users, live data need to be processed, interpreted, filtered and archivedoften using stored data. Archival storage of past sensor data requires a storage system. A good storage systemmust address issues such as where the data is stored, whether the data is indexed and how the application canaccess this data in an energy efficient manner. One such storage system, Two-Tier Storage Architecture(TSAR), is analyzed in this section. TSAR is a two tier storage system which seeks to improve upon theexisting homogenous storage system. The evaluation of TSAR was presented in the paper "TSAR: A Two TierStorage Architecture Using Interval Skip Graphs" [Desnoyers05].2.1 Experimental Set-upPerformance Analysis of Wireless Sensor Networkshttp://www.cse.wustl.edu/~jain/cse567-06/ftp/sensor_perf/index.html 2 of 12An emulator is employed to evaluate the performance of lookup and update overheads using real andsynthetic datasets. The performances of the factors are evaluated by graphing the number of messages sentagainst the number entries in the network. The only metric of concern is time taken for an operation.2.2 ResultsFigure 1(a): Cost of Insert - Empirical Data Figure 1(b): Cost of Insert - Synthetic DataFigure 1(a) and 1(b) show the cost of insert using the real data and synthetic data respectively. Each insertentails an initial traversal followed by neighbor pointer updates. The graphs show that the initial traversalincurs a cost of log n and the neighbor update incurs a cost of 4 log n.Figure 2(a): Cost of Lookup - Empirical Data Figure 2(b): Cost of Lookup - Synthetic DataFigure 2(a) and 2(b) show the cost of lookup using the real data and synthetic data respectively. Lookupinvolves first, looking up the first interval that matches the query and, in the case of overlapping intervals,traversing to identify all matching intervals. It can be inferred from the graphs that the initial lookup takes logn however the subsequent traversal is largely dependent on the size of the data.2.3 CritiqueVerifying the expected performance of insert and lookup is good because the performance of the storagesystem largely rests on the operations it provides. However, the set of operations evaluated is not complete.Other important operations such as delete are not evaluated at the same depth as insert and lookup.2.3.1 DataRetrospective data from the real world does not mean duplication of the real world. The real world offersinfinitely many possibilities so there are bound to be scenarios that are omitted. Having said that, the use realsynthetic data is appropriate in this case as the author is trying to quantify the performance of factors. Thedata set is also large enough to represent a real world scenario.Performance Analysis of Wireless Sensor Networkshttp://www.cse.wustl.edu/~jain/cse567-06/ftp/sensor_perf/index.html 3 of 122.3.2 WorkloadsNo details are given on the selection of the workloads. All that is communicated in the paper is the nature ofthe data collected from the workload - empirical and synthetic.2.3.3 FactorsFactors evaluated are insert and lookup operations. Delete, another important factor, is not evaluated althoughthe author promised to evaluate this factor. It could be argued that the performance of delete may becomparable to insert. Nevertheless, the author should have evaluated delete based on special operations thatdelete might have.2.3.4 MetricsNumber of messages sent is an appropriate metric for evaluating at the user's end but not for the programmer.The programmer will need more details to know where the
View Full Document
Unlocking...