2.5 Research Prototypes
2.5.3 IoTSE Form and Implementation Analysis
Comparison Result
The mapping of selected prototypes into dimensions defined in our analytical framework is presented in Figure 2.7, 2.8 and 2.9.
Operating scope of an IoTSE is presented in form of[DiscoveryScope]−[SearchScope].
Local scope denotes that the IoTSE can only find resources and provide search services to users in its vicinity, whileGlobal scope denotes that it can operate across the Globe via the Web. An IoTSE can be tailored to work withHuman orMachine users. It’s evaluation can be carried withPrototypes on real devices orSimulation (e.g., network simulation with NS2).
The scheme of discovery process carried out by an IoTSE can be either Active or Passive. Active discovery means the search engine seek resources, while Passive discovery means resources are registered to the search engine. Depending on the scope and discovery scheme, an IoTSE uses different types of collector, including Web Crawlers, resource Registration mechanisms and Local Discovery (LD), which includes mechanisms to detect entities and resources in the immediate vicinity. The support for mobile objects by an IoTSE is organized into four groups. Timer denotes the continuous resampling of object’s location after a predefined time period. Beacon denotes the mechanism in which the search engine continuously broadcasts beacons for receiving objects to register themselves. Ad-hoc Pull (AHP) denotes that the location of objects are pulled every time a query is processed. Mobile
2.5 Research Prototypes 41
Query:
Query Processing in WoTSE:
Meta-Path:
Meeting Room 1
Meeting Room 2
Auditorium 1 Resource Content
Sensor 1 “Unoccupied”
Sensor 2 “Occupied”
Sensor 3 “28 Celsius”
Sensor 4 “Unoccupied”
Sensor 5 “32 Celsius”
Resource Metadata Rep 1 “type : meeting”
Rep 2 “type : meeting”
Rep 3 “type: auditorium”
Rep 1 “type : meeting” Home Page
Result:
D(Content) + R(Metadata) => Object => R
“Find an available meeting room in the building”
“Find representative Web Pageof a room object with the type called “meeting” and the real-time state reported as unoccupied”
Discovered Sensor Stream Resources
Discovered Representative
Resources Link between an object and a resource
Fig. 2.3 Assessment of a query for available meeting room in a smart building and its related meta-path
Web of Things:
Sensor Streams Representatives Functionality
Websites Web Services
Search Engine:
Users:
Interfaces:
Search:
Discovery:
Index:
Security, Privacy and Trust
Application Human
Discoverer Retriever
Storage
Manager Indexer Q.I Ranker
Resource
Collections Indexes Q.I Ranking Score
Q.D Ranker Rank
Aggregator Query
Processor
Result Processor Query
Interface
Result Interface
Fig. 2.4 A Modular Architecture for Web of Things Search Engines
2.5 Research Prototypes 43
DBLP
Scopus
Meta-path Scope
Meta-path (4.1)
Modular Architecture (4.2)
Discovery Dimensions
Index Dimensions
Search Dimensions
User Interface Dimensions
Security, Privacy,
Trust
Experiment
Selection Algorithm
Related Works
Prominent Subset
Detailed Analysis
Direction &
Implementation (6.3) Growth of the
Field (6.2) Results:
Data Collecting:
Modelling:
Publication Analysis Analysis:
Fig. 2.5 Overview of the analytical framework. Oval objects represent components that we created. Dash arrows denote that the pointed object is derived from the pointing object. Solid arrows represent represents the link between inputs and outputs of our analysis.
2001 2002 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016
Number of Publications 2 1 5 6 4 11 7 11 15 23 29 38 38 23
Number of Cited Works 2 1 5 5 4 9 5 10 8 11 10 6 5 0
Sum of citation 19 5 23 18 19 31 18 67 42 34 17 9 9 0
0 10 20 30 40 50 60 70 80
Publication vs Citation
Fig. 2.6 Number of publications, in-field cited works and in-field citations
PublicationMeta-pathScopeExperiment TypeExperiment Scale Max (Yap, et al. 05)2005R(Con) => E => R + DL-GProt10 participants GSN (Aberer, et al. 07)2007D(ID) => DL-G-- SenseWeb (Kansal, et al. 07)2007D(Meta) => DG-G-- DIS (Yan, et al. 08)2008D(Con) => E => RL-GProt6 sensors Microsearch (Tan, et al. 2008)2008S(Con) => SL-LProt1 sensor OCH (Frank, et al. 2008)2008R(ID) => E => R + DL-GProt,Sim4 participants Dyser (Ostermaier, et al. 10)2010D(Con) + R(Con) => E => RG-GSim385 sensors uKB (Ruta, et al. 10)2010R(Con) => RL-GSim50 readers Snoogle (Wang, et al. 10)2010R(Con) => E => R + DL-GProt8 sensors WoT Discovery (Mayer, et al. 11)2011R(Con) => RG-G-- IteMinder (Komatsuzaki, et al. 11)2011R(Con) => E => R + DL-GProt1 robot Christophe, et al. 112011F(Meta) => FG-GProt- Mathew, et al. 112011R(Con) => RL-G-- WoT Search (Mayer, et al. 12)2012D(Con) + R(Con) => E => RL-GSim600 sensors Wei, et al. 122012F(Meta) => FG-G--
PublicationMeta-pathScopeExperiment TypeExperiment Scale CASSARAM (Penera, et al. 13)2013D(Meta) => DL-GSim- Content-based (Truong, et al. 13)2013D(Con) => E => RG-GProt,Sim163 sensors Ambient Ocean(Carlson, et al. 14)2014R(Meta) => RG-GSim42,000 signal samples (Mrissa, et al. 14)2014F(Meta) => FL-L-- IoT-SVK (Ding, et al. 14)2014D(Con) + R(Con) => E => RG-GProt10,000 Taxis Gander (Michel, et al. 14)2014D(Con) + R(Con) => E => RL-GProt63 devices DNS (Kamilaris, et al. 14)2014R(ID) => E => FG-GSim100 sensors AntClust (Ebrahimi, et al. 15)2015D(Meta) => DG-GSim100,000 sensors ForwarDS-IoT (Gomes, et al. 15)2015R(Meta) => RG-GSim- ECS (Shemshadi, et al. 15)2015D(Con) + R(Con) => E => RG-GSim10,000 Taxis Renna, et al. 162016R(Con) => RG-GProt- Chen, et al. 20162016R(Meta) => RG-GProt8 participants VisIoT (Nunes, et al. 16)2016D(Meta) => DG-GSim- ThinkSeek (Shemshadi, et al. 16)2016D(Con) + R(Con) => E => RG-GProt- LHPM (Zhang, et al. 16)2016D(Con) => E => RL-GSim132 sensors Fig.2.7Meta-pathTypes,ScopeandExperimentScaleofselectedPrototypes.
2.5 Research Prototypes 45
Discovery SchemeMobility SupportCollector TypeCollection TypeIndex TypeQ.I RankingStorage ScalabilitySearch SchemeQuery ModelResult ModelQ.D RankingAdaptabilitySearch ScalabilityUser TypeInterface ModalQuery InterfaceResult InterfaceSecurityPrivacyTrust Max (Yap, et al. 05)AAHPLDV--VAHTxtLD(Rnk)--HWTBxL-OAC,LAC- GSN (Aberer, et al. 07)ATLDR---AHIDStrExt--M---CrptOAC- SenseWeb (Kansal, et al. 07)PMPRRU/S--AHCondStrD(Rnk)-CchMAPIAPI--OAC, SumR DIS (Yan, et al. 08)P-RRTxt-DAH+CTxtLD(Rnk)-DHApp-L--- Microsearch (Tan, et al. 2008)P-RRTxt--AHTxtLD(Rnk)--HAppTBxL--- OCH (Frank, et al. 2008)AAHPLDV--VAHIDSExt-ScpHAppImpS-OAC- Dyser (Ostermaier, et al. 10)A-CrwlRPM--AHTxt+ CondLD(Rnk)+ Ext-ScpHWTBxL--- uKB (Ruta, et al. 10)AAHPLDV--VAHCondLExt--M------ Snoogle (Wang, et al. 10)PT+BRRTxt-DAHTxtLD(Rnk)-DHAppTBxLCrptOAC- WoT Discovery (Mayer, et al. 11)P-RV--VAHIDSExt--MAPIAPIS--- IteMinder (Komatsuzaki, et al. 11)A-LDR---AHTxtSExt--HAppTBxM--- Christophe, et al. 11P-RRU/S--AHCondLD(Rnk)ReqT-H+MAppImp---- Mathew, et al. 11P-RR---AHCondLExt--HAppTBxL--- WoT Search (Mayer, et al. 12)AAHPLDR--DAHTxt+ CondLD(Rnk)-DH+MAPIAPIL--- Wei, et al. 12P-RR-QoS-AHCondLExt--MAPIAPIL--- Fig.2.8Implementationofselectedprototypes.“-”denotesthatthefeatureisnotdescribed,while“N/A”denotesthatthefeatureis notimplemented.
Discovery SchemeMobility SupportCollector TypeCollection TypeIndex TypeQ.I RankingStorage ScalabilitySearch SchemeQuery ModelResult ModelQ.D RankingAdaptabilitySearch ScalabilityUser TypeInterface ModalQuery InterfaceResult InterfaceSecurityPrivacyTrust CASSARAM (Penera, et al. 13)A-LDR---AHCondLD(Rnk)+ Ext--H+MApp+APITBx+APIL--- Content-based (Truong, et al. 13)A-CrwlRPM-DAHCondLP(Rnk)-DHWFL--- Ambient Ocean(Carlson, et al. 14)A-CrwlRU/SR-AHTxtLD(Rnk)--HAppTBx+SenL--- (Mrissa, et al. 14)P-RR---AHCondLExt--M----F- IoT-SVK (Ding, et al. 14)PTRegRTxt,S,V-DAHTxt+ CondLExt-DH------ Gander (Michel, et al. 14)AAHPLDV--VAH+CCondLExt-DMAPIAPIL--- DNS (Kamilaris, et al. 14)P-RR--DAHIDLExt-DMAPIAPIL--- AntClust (Ebrahimi, et al. 15)P-RRClst--AHCondLD(Rnk)--MAPIAPIL--- ForwarDS-IoT (Gomes, et al. 15)P-RR--DAH+CCondLExt--H+MApp+APITBx+APIL--- ECS (Shemshadi, et al. 15)ATCrwlRClst--AHTxt+ CondLD(Rnk)--HWTBxM--- Renna, et al. 16---R---AHTxtLD(Rnk)--M------ Chen, et al. 2016---RClst--AHTxt+ CondLD(Rnk)--HAppTBxL--- VisIoT (Nunes, et al. 16)P-RR---AHCondLD(Rnk)--HWFM--- ThinkSeek(Shemshadi, et al. 16)ATCrwlRClst--AHTxt+ CondLD(Rnk)--HWTBxM--- LHPM (Zhang, et al. 16)P-RRPM-DAHCondLP(Rnk)--MAPIAPIL--- Fig.2.9(Cont)Implementationofselectedprototypes.
2.5 Research Prototypes 47 Proxy (MP) denotes the use of spatially deployed proxies to query for resources at specific locations without having to keep track of their mobility [47].
The Collection Type includesReal andVirtual collections. Index Type includes Text- based (Txt) indexes, which also include image-based indexes that treat images as a set of terms [45],Spatial Indexes, numericalValue Indexes,Clustering mechanisms, Prediction Models (PM) (e.g., Sensor Rank [21]) and Unspecified Indexes (U/S) denoting indexing schemes that are mentioned but not described by the prototype. Q.I Ranking dimension in- cludes the use of Quality-of-Service (QoS) andRatings from community. Storage Scalability support includes the use ofVirtual resource collections to negate the need of actual storage andDistribution of resource storage over multiple instances of the search engine.
The Search Scheme dimension includes Ad-hoc (AH) andContinuous search, denoting whether the given queries are matched one time against the current snap shot of the resource collection or continuously assessed against the updating collection. The Query Model dimension includes Text-based queries (Txt), LogicalConditions andIDentity. The Search Result can be a List of matching data records, aSingle record, or a Stream of dynamic information (e.g., sensor readings). Q.D Ranking includes ranking based on the value of prediction models (P(Rnk)), distance-based ranking (D(Rnk)), which can be expressed by Euclidean distance, Jaccard index or Cosine similarity, and exact matching (Ext). We consider Text-based ranking (e.g., TF-IDF) a form of D(Rnk). Search Scalability mechanisms include theDistribution of query processing, caching (Cch) search results to reduce number of query sending to sensors (e.g., SenseWeb [47]) and scoping (Scp) to reduce the number sensors to assess. The adaptability dimension includes only one value - ReqT - which denotes the ability of a search engine to detect and adapt its algorithm to the type of user making the request.
The Interface Modal denotes the channel of communication between a search engine and search users, includingWeb Interface, WebAPIand specializedAPPlication. The form of
interface on this channel to receive queries from users includes structuredForms, text boxes (TBx),Sensors on client device, andImplicit queries invoked by the interaction between users and client application (e.g., [22]). Result interface includes the traditional List of records and the geographicalMap.
Security measures of IoTSE include encryption (Crpt). Privacy in IoTSE is protected by enforcing access control on objects (OAC) and spatial locations (LAC),Summarizing sensor data andFiltering of search results to protect sensitive information of involved users. Finally, on Trust dimension, we have the valueRdenoting the use of ratings from the community.
Form of IoTSE
Figure 2.10(a) presents the distribution of meta-paths supported by the selected prototypes.
Searching for objects based on their ID or metadata (R⇒R) is the most common form of IoTSE, followed closely by searching for objects using their real-time state (e.g., sensor readings, location) and searching for sensor streams (D⇒D). Familiarity is a possible expla- nation for the popularity of these meta-paths. For instance,R⇒Ris similar to Web search, while R+D⇒Ob j⇒Rcomes naturally with the idea of feeding real-world states into software applications. Surprisingly, searching for real-world functionality is not commonly supported even though it is crucial in the interaction with IoT-enabled smart environments.
Figure 2.10(b) presents the distribution of operating scopes of selected prototypes. Local and global resource discovery are equally supported by the prototypes, which reflects the attention to both ends of the IoT scale. However, global scope dominates the search operation.
Implementation of IoTSE
Figure 2.11 presents the support that key dimensions receive from the selected prototypes.
Q.D ranking and discovery enjoy the strongest support, as they are the core of an IoTSE.
Scalability of query processing and storage capability is supported by about half of prototypes.
2.5 Research Prototypes 49
3.33% 3.33%
10.00% 10.00%
13.33%
16.67%
20.00%
23.33%
0.00%
5.00%
10.00%
15.00%
20.00%
25.00%
Distribution of Simplified Meta-paths
(a) Meta-paths
G-G 53%
L-G 40%
L-L 7%
Proportions of operating scopes
(b) Scopes
Fig. 2.10 Distribution of Meta-paths and Operating Scopes
Most supporting prototypes scale up by utilizing virtual resource collections and distributing processing and storage across multiple computers. Mobility of physical objects is considered by less than 40% of the selected prototypes. The weak support for indexing is a surprising result, considering its crucial role in resolving queries. A possible explanation for this phenomenon is the simplicity of usage scenarios and involving resources in the selected prototypes.
Most prototypes do not support adaptability which means that they cannot change their operations according to context, such as their current users. Query Independent (Q.I) ranking also lacks support, even though it plays a crucial role in the success of Web Search engines.
It can be contributed to the lack of natural order of IoT resources. Security, privacy and trust are also not commonly addressed by prototypes.
Figure 2.12 presents the details of some interesting dimensions that receive high support.
On discovery scheme, the active and passive schemes are equally utilized. This is a surprising result because both Web Search and Sensor Search systems, which are frequently considered predecessors of IoTSE, rely on active discovery scheme. On collection type dimension, real collections dominate because it is most straightforward and traditional solution in search engines. Search scheme is dominated by ad-hoc search scheme, which is carried over from Web Search Engines. On targeted user type dimension, human users have a slight edge over
0% 20% 40% 60% 80% 100%
Discovery Scheme Mobility Support Index Type Q.I Ranking Storage Scalability Q.D Ranking Adaptability Search Scalability Security Privacy Trust
Fig. 2.11 Support of Prototypes on key Dimensions
machine users. Interestingly, some IoTSE are designed to support both types of users. On Query Dependent (Q.D) ranking dimension, distance-based ranking and exact matching are two most common forms of ranking mechanisms among the selected prototypes. On the query model dimensions, logical conditions is the most common form of query, while list of resources is the most common result model.