Imagine connecting thousands of web browsers with browser-to-browser connections, sharing storage, bandwidth, and CPU. This builds a fog of browsers where end-user devices are ready to collaborate. Imagine semantic fog applications running in fogs of browsers, querying the linked data servers hosted in the cloud and data hosted in the fog. Fogs of browsers running semantic fog applications create a new massively decentralized infrastructure where RDF data and SPARQL query processing are available both on web servers and on browsers. In this paper, we explore new opportunities and research challenges opened by a fog of browsers for the semantic web.
The paper is organized as follows. Section 2 de nes semantic fog applications in the fog of browsers. Section 3 presents new opportunities opened by semantic fog applications. Section 4 highlights new research challenges for semantic fog applications. Finally, conclusions are outlined in section 5. 2
http://myfog b0
b1
b3 b2 b4 b5 b6 b7 b8 b9
A fog of browsers is a set of interconnected browsers with browser-to-browser connections. Such connections are now supported thanks to the WebRTC standard in Firefox, Chrome, Microsoft Edge and IOS. A browser can participate to one or several fogs.
A fog of browsers is accessible through one or several URIs hosted on regular web server.
The web server dereferences this address to a JavaScript application bootstrapped with a
sample of already connected browsers2. This JavaScript application represents a semantic fog
application with its own logic. The application is able to manage RDF data and runs SPARQL
queries over linked data and/or over data hosted in the fog. We assume that all RDF data
are managed following the linked data principles [
Once downloaded in the browser, the semantic fog application joins the network of browsers by connecting the browser to at least one of the already connected browsers. Following this approach, at a given time, there is potentially a high number of browsers, running the same application, and all these browsers are connected together. We do not make any assumption about the topology of the network, i.e. hierarchical, structured, unstructured, hybrid or multi-layer. Topology depends on the objective of the semantic fog application. In gure 1, the browsers are connected in an unstructured network, they execute SPARQL queries over data in the fog and 2 datasources hosted in the cloud. The browser b0 contacts the web server hosting the semantic fog application in order to join the network. The web server is returning the semantic fog application and references to two browsers: b1 and b2. b0 contacts one of them to join this fog.
To be usable, a semantic fog application must meet the following requirements inspired
from P2P data management [
Deploying semantic fog applications over a fog of browsers raises several opportunities. In this section, we present several semantic fog applications illustrating di erent usages. 3.1
Queries in the fog
Connect b r o w s e r s g e o l o c a l i z e d i n Nantes / Overlay . c o n f i g u r e ( G e o l o c a t i o n =' Nantes ' ) ;
Q1 : Get nearby p o i n t o f i n t e r e s t / S t r i n g query= ' s e l e c t ? p l a c e where ? p l a c e nearby ( '+ m y p o s i t i o n +' 100m) ' ; R e s u l t S e t l o c = q u e r y E x e c u t i o n . e x e S e l e c t ( query , model )
Q2 : C o l l e c t u s e r f e e d b a c k / answer=askUser ( ' Do you l i k e your l o c a t i o n ?: '+ l o c ) UpdateAction . e x e c u t e ( ' INSERT DATA'+ 'me'+op : l i k e s+ l o c ) ; / Q3 : D i s p l a y most l i k e d p l a c e s / q u e r y E x e c u t i o n . e x e S e l e c t ( 'SELECT ? p l a c e COUNT( ? l i k e s ) f ? p l a c e l i k e d ? o g groupby ? p l a c e ' ) )
Consider a simple semantic web application where people visiting a city have access to point of interests around them, can rate these points, and can list top-ranked point of interests. This application can be written with queries like Q1, Q2, and Q3 presented in gure 2 and can be deployed in the cloud. Consequently, queries are executed in the cloud with data stored in the cloud datastore. In this case, the cost of running this application relies entirely on the application provider, the availability and the performances of the application relies on the cloud provider.
Now consider that the code of the Figure 2 is a semantic fog application that is loaded and run in each browser visiting the web page of this application. The line 4, the semantic fog application connects the browser of the visitor to a fog of browsers where browsers are now located in the city of Nantes.
Concerning query Q1, data are still located in the cloud, but now the fog could provide data caching and consequently could improve data availability and reduce the cost of data providers. Under certain conditions, the application could continue to run even if cloud services are unavailable.
Concerning query Q2, the semantic fog application can be con gured to store data locally in the browser, in the fog, or in the cloud. Suppose user feedback is stored locally in the browser. Then the cost of executing Q2 is no more on the charge of the application provider and furthermore, user ratings do not leave browsers.
Query Q3 is dependent on data location. Di erent situations are possibles: if users ratings are stored in the cloud, then executing Q3 in the fog is similar to Q1. If users ratings are stored in their own browsers, then Q3 execution requires to contact every browsers. If users ratings are stored somewhere in the fog, then data can be smartly located and aggregated to answer Q3 e ciently without contacting every browsers.
As we can see, running semantic fog application opens di erent trade-o concerning what can be done in the fog and what can be done in the cloud. This trade-o impacts the cost of running an application, the performance of the application, and the availability of the application. It can also impact the privacy of personal data and the quality of collected data. 3.2
DrugBank
DBpedia Serveur TPF 1
Cache HTTP 1 C1 C1
C2 C2
C3 C3
C4 C4
C5 C5
C6 C6
C7
C8 C7
C8
C9 C9
Cyclades [
Cyclades is based on a double overlay networks; the rst one builds a random network providing connectivity while the second one incorporates a similarity metric. The similarity metric is able to detect users performing similar queries based on the analysis on their local caches. The two-level network topology of Cyclades is described in gure 3.
In this scenario, the fog is able to reduce the number of calls to data providers. Consequently, this improves data availability and reduces the cost of providing data. 3.3
Ladda [
By parallelizing the execution of queries over di erent browsers, the execution time of this workloads can be signi cantly reduced. Figure 4 illustrates a Ladda's query execution. In this execution, a browser executes 1509 queries with the help of 6 neighbors in a network composed of 50 participants. Each square represents the execution time of a query on the swim lane of a browser. On this run, the execution time of the workload is 2m37s instead of 3m32s if the workload was executed by one browser.
In this scenario, the semantic fog application allows to share the CPU and bandwidth of browsers for SPARQL query processing. 4
Deploying semantic fog applications on a fog of browsers opens many opportunities for semantic web application developers. They can optimize nancial costs, availability, performances, privacy . . . However, the programming model has to remain simple as the one depicted in Figure 2. The con guration of the semantic fog application has to determine how queries and data are deployed in the cloud and in the fog to reach developper expectations.
The fog of browsers can reuse some scienti c results from P2P data management
systems [11, chapter 16]. Many works demonstrated how data can be e ciently stored and
accessed on structured, unstructured, and hybrid P2P networks such as Edutella [
The constraints of WebRTC change the cost of communications and potentially impact all existing algorithms.
Customized overlay networks for a fog of browsers. A fog of browsers connects
thousands of browsers over WebRTC. The nature of WebRTC networks and the objective of the
semantic fog application can lead to di erent design choices. As routing is costly in WebRTC,
keeping useful neighbors around us in one hop, can be a good strategy for e ciency and
quality of service. Indeed, direct neighbors can be contacted at low cost. 'useful neighbors' can
have di erent meanings according the application. Many similarity metrics can be de ned and
many overlay can be combined in the same fog as proposed in [
Dynamic replication and consistency in a fog of browsers. Data replication is a fundamental concept for improving data availability and performances of query processing. In the context of a fog of browsers, replication contributes to query e ciency, quality of service and fault-tolerance requirements. A replication strategy has to decide what data to replicate, where to replicate and when to replicate. Such decisions are complex in a fog of browsers: the participants are autonomous, the data storage is limited, the communication costs constrained by network topology. Adaptivity of replication to queries seems a good strategy. Materializing data fragments that are frequently retrieved from data providers and spreading them within the fog can have a signi cant impact on performances. De ning these fragments, deciding when to replicate them and where to locate them is clearly challenging. Another challenge strongly related to data replication is consistency management. Data needs to be up-to-date. Maintaining consistent data fragments at low-cost in a fog of browsers is clearly challenging. Crowdsourcing with a fog of browsers A browser is not just an execution environment for JavaScript programs. It could also involve humans with their Web of Things devices. Fog computing allows a collaboration between man and machines to collect, curate and aggregate data. Consequently, a fog of browsers can be seen as a distributed crowdsourcing platform where data are collected, semanti ed and veri ed within the fog, before saved to the cloud. How the functionalities of a crowdsourcing platform can be distributed among the fog and cloud providers is an interesting challenge.
Federated query engines for a fog of browsers. Federated SPARQL query engines [
Security for semantic fog application If a fog of browsers opens many opportunities, it also brings new threats: A fog of browsers can be used to perform DDOS attacks, to steal personal information from browsers, and to watch people. A semantic fog application has to protect participants and data providers against malicious users. Semantic fog applications require appropriate security models. 5
In this paper, we presented how semantic fog applications running in the fog of browsers creates a massively decentralized infrastructure that extends the semantic web to the browsers of end-users. By this way, the semantic web can take advantage of resources of browsers, including end-users and IoT devices. Semantic fog applications can improve the e ciency and quality of service of linked data providers. It can also enhance the linked data with data provided by end-users.
If some semantic fog applications are already there, more research e orts are needed to fully exploit all the potential of semantic fog applications: pertinent network topologies, dynamic replication, e cient query processing, data quality and security.
Another interesting research questions have not been discussed in this paper: the dynamicity of the fog of browsers and how fog of browsers can be combined with distributed ledgers for commercial query processing in the fog of browsers.