We propose a Context Aware Sensor Con guration Model (CASCoM) to address the challenge of automated context-aware con guration of ltering, fusion, and reasoning mechanisms in IoT middleware according to the problems at hand. We incorporate semantic technologies in solving the above challenges.
Broadly, con guration in IoT paradigm can be categorized into two: sensor-level
con guration and system-level con guration. Sensor-level con guration focuses
on changing a sensor's behaviour by con guring its embedded software
parameters such as sensing schedule, sampling rate, data communication frequency,
communication patterns and protocols. In this paper, we are focused on
developing a system-level con guration model for IoT midddleware platforms.
Specifically, our proposed model identi es, composes, and con gures both sensors and
data processing components according to the user requirements. In existing IoT
middleware (e.g. GSN), many con guration les and programming codes need
to be manually de ned by the users (without any help from GSN). An ideal IoT
middleware con guration model should address all the above mentioned
challenges. The con guration model we propose in this paper is applicable towards
several other emerging paradigms, such as sensing as a service [
In phase 1, users are facilitated with a graphical user interface, which is based on
a question-answer (QA) approach, that allows to express the user requirements.
Users can answer as many question as possible. CASCoM searches and lters the
tasks that the user may want to perform. From the ltered list, users can select
the desired task. The details of the QA approach are presented later in this
section. In phase 2, CASCoM searches for di erent programming components
that allow to generate the data stream required. In phase 3, CASCoM tries to nd
the sensors that can be used to produce the inputs required by the selected data
processing components. If CASCoM fails to produce the data streams required
by the users due to insu cient resources (i.e. unavailability of the sensors),
it will provide advice and recommendations on future sensor deployments in
phase 4. Phase 5 allows the users to capture additional context information.
The additional context information that can be derived using available resources
and knowledge are listed to be selected. In phase 6, users are provided with one
or more solutions3. CASCoM calculates the costs for each solution in using
technique disucced in [
(a) Current GSN Work-flow
Fig. 2. Con guration Execution- ow Comparison: (a) Current GSN (b) CASCoM
Phase 1:
Phase 2:
Phase 6 Phase 5 Phase 4
Applications Con(OOteppxttit-miboiaznsaeatidlo)nC:ost Dis(cOovpCetorinoAtnedxdatilt)io:nal PRreo(cOvoidmpetmiaoednnvdaicale)ti:oannsd Fig. 3. The Context-Aware Sensor Con guration Model (CASCoM) 4 Evaluation, Discussion and Lessons Learned Results: Figure 5(a) shows that CASCoM allows to considerably reduce the time required for con guration of data processing mechanism in IoT middleware. Speci cally, CASCoM allowed the three types of users to complete the given task 50, 80 and 250 times faster (respectively) in comparison to the existing approach. According to Figure 5(b), the Java re ection approach takes slightly more time to specially when initializing. Though the Java re ection approach can add more exibility to our model, the additional overhead increases when the number of components and operation involved gets increased. The overheads can grow up to unacceptable level very quickly when GSN scales up (e.g. more user requests).
According to Figure 5(c), even IT experts who know GSN can save time by using CASCoM up to 88%. Specially, time taken for de ning the VSD and VS class have been signi cantly reduced. Both les can be generated by CASCoM autonomously within a second even for complex scenarios. However, the time taken to nd data processing components and sensors (and wrappers) depends on the size of the semantic data model. Figure 5(d) shows how total processing time would vary depending on the size of the semantic data model. Approximately, a semantic model with 10,000 sensor descriptions and 10,000 data processing components can be processed in order to nd solutions for a given user request in less than a minute. However, most of the time is taken to read the data model. ) c e s ( n o itr a u g if n o c e h t ltpee l)cae mS o ic c m to itrh e ag iTm(Lo
Usecase (1) Different Scenarios
15% 14% 20% 51%
7% 1% 3% 1% 88% (c) Non-IT Expert without GSN Skills (without CASCoM)(a)
(1) (5)
(10) Search Wrappers Search Programming Components Time Saved
Define VSD Write VS Class
0 250N0umber of 5re0c0o0rds mode7lle5d00 Fig. 5. Evaluation of CASCoM 10000 The actual con guration process other than reading the data model takes only 4 seconds and it slightly increases when the model size increases. 5
We have shown that it is possible to o er a sophisticated con guration model to support non-IT experts. Semantic technologies are used extensively to support this model. Using our proof of concept implementation, both IT and non-IT experts were able to con gure the GSN in signi cantly less time. In future, we plan to extend our con guration model into sensor-level. To achieve this, we will develop a model that can be used to con gure sensors autonomously without human intervention in highly dynamic smart environments in the IoT paradigm.