Section: New Results
Service-oriented Computing in the Future Internet
Participants : Georgios Bouloukakis, Nikolaos Georgantas, Valérie Issarny, Ajay Kattepur, Raphael de Aquino Gomes, Rachit Agarwal.
With an increasing number of services and devices interacting in a decentralized manner, choreographies represent a scalable framework for the Future Internet. The service oriented architecture inherent to choreographies allows abstracting diverse systems as application components that interact via standard middleware protocols. However, the heterogeneous nature of such systems leads to choreographies that do not only include conventional services, but also sensor-actuator networks, databases and service feeds. We reason about the behavior of such systems by introducing abstract middleware connectors that follow base interaction paradigms, such as client-service (CS), publish-subscribe (PS) and tuple space (TS). These heterogeneous connectors are made interoperable through a service bus connector, the eXtensible Service Bus (XSB) [11] .
In previous work, we identified and verified the behavioral semantics of the XSB connector derived from the interconnection of base connectors, and introduced a method for constructing protocol converters enabling this interconnection. We implemented our XSB solution into an extensible development and execution platform for application and middleware designers. We also provided a lightweight implementation of the XSB, the Light Service Bus (LSB), appropriate for resource-constrained environments and systems. Next, leveraging on the functional interoperability across interaction paradigms offered by the XSB, we initiated our study of end-to-end Quality of Service (QoS) properties of choreographies, where in particular we focus on the effect of middleware interactions on QoS.
Building on the above results, we refine our analysis of QoS on top of the identified interaction paradigms. We have introduced a motivating application scenario inspired from the 2014 D4D Challenge (http://www.d4d.orange.com/en/home ). More specifically, Data for Development Senegal is an innovation challenge on ICT Big Data for the purposes of societal development. Mobile network provider Sonatel (part of the Orange Group) has made anonymous data extracted from the mobile network in Senegal available to international research laboratories, encouraging research related to the development and welfare of the local population.
Our scenario targets the development of an application platform for citywide and countrywide transport information management relying on mobile social crowd-sensing. This takes into account the particular context and constraints in Senegal. More specifically, the local transportation system, although developing, still consists of many unplanned and informal settlements with unreliable services and infrastructure. Additionally, despite wide use of mobile phones in the country, mobile Internet access remains limited, making SMS the only alternative for data access for a large part of the population. Our proposition aims to complement the scarce authoritative transport information coming from structured information sources and compensate for the lack of such information. In particular, in our approach we intend to study and experiment with appropriate interaction paradigms (CS, PS, TS) on top of 3G/2G/SMS data connections, further depending on the specific application and data. We are especially interested in interaction adaptation depending on the network conditions (e.g., switching to SMS-based protocol when the 3G/2G network is unavailable).
We have taken a first step towards enabling such an application platform. This consists in evaluating the publish/subscribe interaction style in a large-scale setting where resources of mobile users are limited, which translates into limited and intermittent connectivity in the system. Additionally, such an application platform must guarantee that the sensing data is processed and delivered to the corresponding mobile users on-time, despite the intermittent connectivity of the latter. We have opted for the publish/subscribe paradigm, as it is deemed appropriate for loose spatio-temporal interaction between mobile entities.
In particular, we introduce a queueing network model for the end-to-end interaction within a large-scale mobile publish/subscribe system. We leverage the D4D dataset provided by Orange Labs to parametrize this model. We then develop a simulator named MobileJINQS (http://xsb.inria.fr/d4d#mobilejinqs ) that implements our model and uses the dataset traces as realistic input load to the system model over the time span of a whole year. Prior to this, we extensively analyze the D4D dataset in order to identify the data that we are interested in and infer primary results (http://xsb.inria.fr/d4d ). Based on the results of our simulation-based experiments, we thoroughly evaluate the behavior of the publish/subscribe system and identify ways of tuning the system parameters in order to satisfy certain design requirements. More precisely, we provide results of simulations of our publish/subscribe system with varied incoming loads, service delays and event lifetime periods. We use connection data of various pairs of mobile network antennas to derive realistic traces for both incoming loads and service delays. System or application designers are able to tune the system by selecting appropriate lifetime periods. We demonstrate that varying incoming loads and service delays have a significant effect on response time. By properly setting event lifetime spans, designers can best deal with the tradeoff between freshness of information and information delivery success rates. Still, both of these properties are highly dependent on the dynamic correlation of the event input flow and delivery flow processes, which are intrinsically decoupled.
Our future work includes comparison of the publish/subscribe interaction paradigm with other interaction paradigms (client-server, tuple space), in relation with the network access capacity and the application requirements. Also, we intend to study the response time and success rate for the various combinations of antennas in more fine-grained scales (e.g., check what their evolution is over one day).