Mobile Netw Appl
DOI10.1007/s11036-013-0479-2
Continuously Changing Information on a Global Scale and its Impact for the Internet-of-Things
Stefan Forsstr¨o m·Theo Kanter
©Springer Science+Business Media New York2013
Abstract This article analyzes the challenges of support-ing continual changes of context information in Internet-of-Things applications.The applications require a constant flow of continuously changing information from nsor bad sources in order to ensure a high quality-of-experience.However,an uncontrolled flow between sources and sinks on a global scale wastes resources,such as computational power,communication bandwidth,and battery time.In respon to the challenges we prent a general approach which focus on four layers where we provide a propod solution to each layer.We have realized the general model into a proof-of-concept imple-mentation running on devices with limited resources, where we can moderate the information exchange bad on relevance and sought after quality-of-experience by the applications.In conclusion,we evaluate our solu-tion and
prent a summary of our experiences regarding the impact of continuously changing information on the Internet-of-Things.
内蒙焖面Keywords Pervasive·Context awareness·Real-time·Internet-of-things
This work has been supported by grant2010-00681of VINNOV A the Swedish Governmental Agency for Innovation Systems,and
by grant00163383of the EU European Regional Development Fund,Mellersta Norrland,Sweden.
S.Forsstr¨o m( )
Department of Information and Communication Systems,
Mid Sweden University,SE-85170Sundsvall,Sweden
e-mail:stefan.forsstrom@miun.
T.Kanter
Department of Computer and System Sciences,
七一活动
Stockholm University,SE-16440Kista,Sweden 1Introduction
Today there is a significant proliferation in situation-aware computing,in which different applications and devices can change their behavior depending on the situation of their ur.The applications are integrated into people’s every-day lives and are made aware of their ur’s situation and context,in order to change their own application behav-ior.This opens up a new field of ur friendly rvices, which are able to pervasively adapt per person in order to provide the best possible rvice for that particular ur. The applications do however require a constant feed of continuously changing information from ubiquitous infor-mation sources,in order to make real-time critical decisions making.The sources are often bad on nsors which can provide helpful information regarding the ur’s situation, such as information from the environment in the ur’s sur-roundings.In detail,this has enabled us to u the features of mobile computing and intelligent reasoning in new forms of applications[1],such as nsing campaigns and social interactions in large-scale populations.This also extends to the vision of an Internet-of-Things[2],where many small devices will be ubiquitously connected and communicating with each other.For example gathering nsor information from smartphones,smart homes,and wireless nsor net-works to create new types of context-aware rvices.The current expectations are of the order of50billion connected devices to the Internet by2020[3].
Current Internet-of-Things applications focus on quite narrow scenarios,where they are able to manage the com-munication without scalability issues.This has led to a large proliferation in the area,but it has obscured the real problems associated with large scale deployment.The early success include,for example,health care solutions [4]which can share simple data such as heart-rate
and
Mobile Netw Appl
blood pressure a few times per cond,and location-bad rvices[5]which often claim to be operating in real-time even when they only update the location every few conds. Smart home solutions such as[6]focus on local exchange and reasoning of contextual information,but have limited potential becau they do not share information with all con-nected entities on a global Internet-of-Things.We foree that the future will demand better rvices in the form of faster update frequencies,awareness,and reasoning.Thus, applications will require a continuous stream of nsor
and actuator updates for their regular operation.Hence,as our surroundings become more aware by means of an increas-ing number of nsors,the amount of information on the Internet-of-Things will also increa.
In respon to the earlier shortcomings in relation to the handling of this vast amount of information,we are rearching a general model that is able to widen the appli-cation focus and support continuously changing information from large numbers of sources on the Internet-of-Things. This had led to the identification of the following challenges which must be solved before it is possible to move from the current narrow applications on the Internet-of-Things,to large scale proliferation.
1.The Internet-of-Things is bad on the idea of many
small connected entities which have quite limited capacity on their own,but which together are able to form a usable collaborative collective knowledge.
However,creating this knowledge is expensive in terms of communication and data processing.Therefore the first challenge is to determine mechanisms that adap-tively minimize the resource consumption bad on application demands,which should ensure a good quality-of-experie
nce for the end ur and longevity for the connected entities.
2.This collective knowledge will require data from widely
spread sources,which will be globally distributed.
Hence,this large and constant flow of information in the system renders traditional approaches unsuitable for the Internet-of-Things.Therefore,the cond challenge is to determine mechanisms which are able to manage global data flows in a properly scaling manner,avoid-ing choke points,and which do not expo any central points of failure.
3.Once the raw data has been acquired,it must be
possible to create a higher form of knowledge from it.This demands an information model which is capa-ble of providing knowledge fusion from the data and being simultaneously continuously updated with the latest values.Therefore,the third challenge is in discov-ering an information model that can create knowledge from continuously changing raw nsor values from the Internet-of-Things.4.Becau of the inconsistency of the knowledge and
the continuous feed of raw data from the Internet-of-Things,most knowledge becomes temporal and
tran-sient.Becau of this,queries related to the data might become invalid even before the query has pasd through the whole datat and returned an answer.
樱桃树怎么种植Therefore,the fourth challenge is in determining knowl-edge access models that are able to operate,understand, and make u of the continuously changing knowledge ba.
The remainder of this paper is organized in the follow-ing order:Section2prents the background and impor-tant prior work which has been conducted within the area.Section3prents our general model for enabling an Internet-of-Things which allows continuously chang-ing information.Section4evaluates our approach bad on different types of architectures.Section5prents our proof-of-concept implementation and Section6an evalua-tion thereof.Section7prents our analysis of the impact of continuously changing context information from the Internet-of-Things,and Section8prents our conclusions and future work.
2Background
挑螃蟹的技巧
The idea of distributing small nsor motes in our surroundings,in order to pervasively n our environment has existed for some time.The area has been extensively rearched[7],but the focus has mainly been on the short range communication between the motes and the optimiza-tion thereof.
Thus,the scope of the rearch area has been quite narrow and focud on the hardware technical aspects. This has,in turn,spawned many nsor mote platforms and operating systems which are particularly aimed at wireless nsor networks,such as Contiki and TinyOS.The rearch into Machine-to-Machine(M2M)communication has also proliferated from the pervasive nsing area[8].In this ca, the M2M communication focus is on creating intelligence between different devices,without human interaction.For example changing the houhold heating bad solely on different nsor and actuators in the home involves commu-nication between each of them.However,they often rely on different types of nsor gateways or centralized databas to relay the nsor information on the Internet.Thus,this involves hiding the actual nsor motes in order to protect them from true ubiquitous access and global information sharing.
The concept of utilizing information from nsors attached to different entities,in order to provide more personalized,automatized,or even intelligent application behavior can be referred to as the Internet-of-Things[2].
Mobile Netw Appl
秋风起兮木叶飞This also includes the ideas of having nsors every-where in our surroundings,to form a truly perva
sive and ubiquitous smart space.The reasoning is that everyday objects will become connected and that they can display intelligent behavior.New forms of applications will also become widespread when all of the objects offer different types of rvices.The types of Internet-of-Things applica-tions are projected to have a significant impact with regards to how we interact with the world,people,and things in the future.We believe that the main differences between Internet-of-Things applications and traditional nsor sys-tems are the focus on global scale and the integration into everyday objects.
The large number of nsors and actuators on the Internet-of-Things will create even larger amounts of data that is required to be disminated to applications.Others have also identified this nsor data overload from mobile devices[9],and have indicated the requirement for better scaling infrastructures in the future.Parallels can also be drawn to rearch conducted within the Big Data area[10], which focus on the problems associated with managing and querying large amounts of data.In the light of this, we have performed an initial measurement on the mobile devices available today.We tested a typical mobile phone (a Samsung Galaxy Nexus)and it can produce up to1200 values each cond.Even if nsor values are quite small in size this number is staggering large,especially when viewed in the perspective of sharing nsor information on a global scale.This
can be en as an indication of the impending information overload from nsor sources,which will be a problem for the future Internet-of-Things where all devices will be connected and want to ubiquitously exchange nsor information.
3Propod general model
From the four challenges in Section1,we have derived a layered general model of the areas for which the chal-lenges are expod.This layered general model can be en in Fig.1,where we can e that the fourth chal-lenge regarding knowledge access is the topmost layer and is furthest toward the application.This layer will deal with the challenges regarding applications which are querying data and requesting data derived from the Internet-of-Things.The cond layer is the information model, which deals with the challenges regarding the organiza-tion of data and how to store it in an efficient manner with regards to each connected thing.The third layer deals with the communication architecture,which must organize the connected things and enable communication between the end points in an efficient manner.Finally,the fourth layer deals with the limited resources available
and Fig.1Propod general model
the actual hardware upon which the information is being produced.
3.1Knowledge access layer
The knowledge access is the topmost layer in our general model,where challenge4regarding accessing information is dealt with.This layer should provide query models that can operate,understand,and make u of the continuously changing temporal knowledge that will come f
红姑rom Internet-of-Things sources.As previously stated,query results might become invalid even before the query has been returned from the information model in the layer below.In detail, we have discovered that the knowledge access layer gener-ally deals with five common problems,namely reliability of a query from the information model(that it returns a usable result),rapid respon(so that the application will not be kept waiting),arching for data(for finding new data sources),two way data flow(for controlling actuators), and authenticated cured data access(to limit the access of private information).
Applications which are accessing knowledge are often conducted through different Application Programming Interfaces(APIs).Even though there are standardized pat-terns with regards to how the APIs should be formed, a vast number of different approaches still exists.In distributed environments publish/subscribe methods have traditionally been ud to solve application information access.This is becau an application has the ability to subscribe to a specific piece of data and receive updates whenever the occur.Lately there have also been a pro-liferation of cloud computing and mashup rvices,which are usually HTTP and REST[11]bad information access. For relational databas the SQL language has been ud for many years in relation to solving application information access,even though different NoSQL[12]databas have gained ground during recent years becau of their ability in relation to handling large amounts of distributed data.
Mobile Netw Appl
3.2Information model layer
The information model is the structure and organization of the data,thus it must deal with challenge3which involves information and knowledge.The application access layer will query the information that an application both desires and requires for its operation.Hence,the information model must solve the demands from the layer above and all of its requirements.The information model is also dependent on the data itlf,and thus places demands on the data that it stores.In detail,we have determined that the information model deals with the following common problems,namely, linking data with other data(and thus traversing it),support for fuzzy values and units(becau not all nsors pro-vide discrete values and are of the same type),and handling continuously changing data from remote sources(becau nsor values are constantly changing).
The information model is the focus of many publications with regards to architectures for situation aware applica-tions,becau it solves the majority of challenges bad on the stored information.The information model must provide the means in relation to reasoning,linking,arch-ing,relations,comparing,inrting,and the deletion of information.The most common approach within t
he area of Internet-of-Things is to apply some form of ontology, such as the Web Ontology Language OWL[13].However, many more exists[14].In contrast to the more complex information models,quite simple XML schemes such as SensorML[15]also exists,which simply markup the data and allow the applications to create the knowledge from the raw data.
3.3Communication layer
性知识科普The communication architecture enables the exchange of information between the connected entities.Hence,the communication layer is ud by the information model to retrieve and nd information to other devices when they require a particular piece of data.Therefore,this layer must deal with challenge2,which involves continuous data flows.Becau of the inherent structure of the Internet-of-Things we have determined that the communication architecture must deal with the following problems,namely, to avoid unnecessary data exchange(data which is not demanded by the applications),to avoid unnecessary prox-ying of data(to enable effective distribution),to avoid any central point of failure(to increa system resilience),and to have a low overhead(to save resources overall).The recent advances in the Internet-of-Things area has also produced a large number of different general platforms which are made for creating applications on.The general platforms can be organized into three distinct categories,depending on where they store the actual information.In d
etail,the three categories are:centralized,mi distributed,and fully distributed architectures.
Centralized systems store the information under a single administrative authority,either in a single large databa or replicated in a cloud bad manner.Examples of centralized storage include SenWeb[16],SENSEI[17], and SERENOA[18].However,all of the centralized sys-tems have scalability issues when the number of updates and queries increas in magnitude.Therefore,they will have problems associated with support for continuously updat-ing data with low latencies.The systems are also prone to failure,becau they expo single points of failure.
Semi-distributed systems store the information locally on each entity in a peer-to-peer manner,but still maintain a centralized authority for the exchange between peers.The systems u ssion-establishment protocols to exchange the context,but under the supervision of a centralized authority.Mobilife[19],CONTEXT[20],ADAMANTIUM [21],and ETSI TISPAN[22]which is bad on3GPP IMS [23],are examples of such systems.The systems will scale in a better manner when compared with the centralized sys-tems,but they still maintain a centralized component.This will become a bottleneck for the exchange,when the entities perform long queries on a large and continually changing datat,since the centralized component has to administer all the ssions,even if the actual exchange is sometimes performed outside the centralized component.
Fully distributed systems both store and administer the information locally on each entity in a peer-to-peer manner. The systems often utilize distributed hash tables to enable logarithmic scaling when the number of entities increas in magnitude.Examples of such systems are MediaSen[24], SOFIA[25],and COSMOS[26].Naturally,the systems do not contain any single point of failure and are thus more resilient,even if the distribution itlf often requires addi-tional overhead in order to maintain an overlay.The main problem associated with fully distributed systems is that it places a larger responsibility on the end devices.However the end devices can be limited in bandwidth and processing power,which will cau there to be issues when sharing context information on a large scale.
3.4Sensor and device layer
On the Internet-of-Things many different nsors and devices will be connected with different resources and this must be handled in the lowest layer.Currently,a wide range of different hardware exists that are suitable for the Internet-of-Things,for example,environmental n-sors,smartphones,wireless nsor motes,and smart dust. However,the idea of the Internet-of-Things is to connect all of the and to enable collaboration between them,even if they have different ba conditions which are dependent
Mobile Netw Appl
on the hardware.One of the vital parts of the Internet-of-Things is the incorporation of nsors and actuators into everyday objects.Hence,the nsor and device layer must deal with challenge1regarding the limited resources and heterogeneous hardware.As previously stated,rearch into this aspect has been conducted for quite some time,namely in the area of Wireless Sensor Networks(WSN)[7].In WSN,the main focus has always been on the resource lim-itations of the devices,becau of their limited battery capacity and processing power.At the prent time quite feature rich but still resource limited platforms exists,such as IPv6capable TinyOs and Contiki.Many small computer platforms also exists,such as Raspberry Pi[27],which have been ud in some Internet-of-Things concepts.It can also be argued that today’s smartphone platforms also fit into the same resource limited device category,becau of their limited battery capacity and processing power.Although there is a drastic increa in performance as new models of smartphones are being relead.
4Realized model
From the general model and the available related work,we have derived a realized model regarding 望洞庭湖赠张丞相原文
how we believe that the Internet-of-Things should be realized in order to address the four challenges.Figure2prents our real-ized model which shows our propod approach to create a device and nsor agnostic architecture,while applying a fully distributed communication architecture,an infor-mation model bad on contextual entities and application access that provides contextual views of the transient data in the information model.It is important to note that our propod general model and to some extent our realized model,is similar to the ETSI M2M architecture[28]espe-cially in the lower layers(communication and hardware layer).But our approach and theirs differ,becau in our solution we include for example the topmost layers to a larger extent(Application,Knowledge access,and Informa-tion model).They also differ in aspects when it comes to the actual realization,we for example propo a solution to limit the exchange of nsor information in order to handle the dismination of continuously changing nsor values. Furthermore,we propo the usage a of fully distributed communication architecture between the nodes,where the ETSI M2M architecture propos mi distributed solutions bad on IMS.
4.1Sensors,actuators,and devices
We believe that the nsors,actuators,and devices will still be very heterogeneous in terms of capacity and resources,but that they will converge toward an all IP bad framework with a lightweig
ht REST bad protocol such COAP[29]for retrieving nsor values.The first steps toward this are already visible,becau both TinyOS and Contiki support IPv6communication and the COAP protocol for data exchange.However,depending on how powerful the devices are,there will still be a requirement for some type of super device,which can act for the weak nsor mote inside the communication architecture.Thus, we will have aggregation points that are able to commu-nicate the information forward when the resources on the nsors or actuators are limited.In detail,we have realized the hardware layer in two parts,parating devices which can manage the upward layers and devices which are unable
Fig.2Realized
model
