Section: New Results

Revisiting the Abstractions of Service Oriented Computing for the Future Internet

Participants : Mohammad Ashiqur Rahaman, Dionysis Athanasopoulos, Sandrine Beauche, Nebil Ben Mabrouk, Nikolaos Georgantas, Valérie Issarny.

A software architecture style characterizes, via a set of abstractions, the types of: components (i.e., units of computation or data stores), connectors (i.e., interaction protocols) and possibly configurations (i.e., system structures) that serve to build a given class of systems. As such, the definition of a software architectural style is central toward eliciting appropriate design, development and runtime support for any family of systems. The service oriented architecture style may then be briefly defined as follows: (1) components map to services, which may be refined into consumer, producer or prosumer services; (2) connectors map to traditional client-service interaction protocols; and (3) configurations map to compositions of services through (service-oriented) connectors, e.g., choreography and orchestration structures. While the service-oriented architecture style is well suited to support the development of Internet-based distributed systems, it is largely challenged by the Future Internet that poses new demands in terms of sustaining ities such as scalability, heterogeneity, mobility, awareness & adaptability that come in extreme degrees compared to the current Internet. Therefore, we have been working on eliciting software architectural abstractions for the Future Internet by building upon the service-oriented architecture style, as well as on applying them to system design, development and execution.

Complex distributed applications in the Future Internet will be to a large extent based on the open integration of extremely heterogeneous systems, such as lightweight embedded systems (e.g., sensors, actuators and networks of them), mobile systems (e.g., smartphone applications), and resource-rich IT systems (e.g., systems hosted on enterprise servers and Cloud infrastructures). These heterogeneous system domains differ significantly in terms of interaction paradigms, communication protocols, and data representation models, provided by supporting middleware platforms. Specifically considering interaction paradigms, the client/server (CS), publish/subscribe (PS), and tuple space (TS) paradigms are among the most widely employed ones today, with numerous related middleware platforms. In light of the above, we have aimed at eliciting abstractions that (i) leverage the diversity of interaction paradigms associated with today’s and future complex distributed systems, as well as (ii) enable cross-paradigm interaction to sustain interoperability in the highly heterogeneous Future Internet [19] .

Existing cross-domain interoperability efforts are based on bridging communication protocols, wrapping systems behind standard technology interfaces, and/or providing common API abstractions. In particular, such techniques have been applied by the two widely established system integration paradigms, that is, service oriented architecture (SOA) and enterprise service bus (ESB). However, state of the art interoperability efforts do not or only poorly address interaction paradigm interoperability. Indeed, systems integrated via SOA and ESB solutions have their interaction semantics transformed to the CS paradigm. Then, potential loss of interaction semantics can result in suboptimal or even problematic system integration. To overcome the limitation of today’s ESB-based connectors for cross-domain interoperability in the Future Internet, we introduce a new connector type, called GA connector, which stands for “Generic Application connector”. The proposed connector type is based on the service bus paradigm in that it achieves bridging across heterogeneous connector types. However, the behavior of the GA connector type differs from that of classical ESB connectors by bridging protocols across heterogeneous paradigms, which is further realized by paying special attention to the preservation of the semantics of the composed protocols. Indeed, the GA connector type is based on the abstraction and semantic-preserving merging of the common high-level semantics of base interaction paradigms.

Eliciting Interaction Paradigm Abstractions: We introduce a systematic abstraction of interaction paradigms with the following features:

• First, we introduce base CS, PS and TS connector types, which formally characterize today’s core interaction paradigms. The proposed types comprehensively cover the essential semantics of the considered paradigms, based on a thorough survey of the related literature and representative middleware instances.

• Then, we further abstract these connector types into a single higher-level one, the GA connector type. GA is a comprehensive connector type based on the abstract union of CS, PS, and TS, where precise identification of the commonalities or similarities between the latter has enabled the optimization of the former. Further, GA preserves by construction the semantics of CS, PS, and TS.

• In more detail, connector types are formally specified in terms of: (i) their API (Application Programming Interface), and (ii) their roles, i.e., the semantics of interaction of the connected component(s) with the environment via the connector. Regarding the latter, the behavioral specification of roles from a middleware perspective relates to specifying the production and consumption of information in the network, while the semantics of the information are abstracted and dealt with at the application layer. The behaviors of the connector roles are then specified using Labeled Transition Systems (LTS). We precisely define the mapping of the roles implemented by the base connector types to/from the corresponding roles of the GA connector type.

• For both the above abstraction transformations, we provide counterpart concretizations, which enable transforming GA connector primitives to CS, PS, or TS connector primitives and then to concrete middleware platforms primitives.

• Furthermore, based on the GA abstraction, we introduce mapping transformations between any pair from the set {CS, PS, TS} via GA. The fine knowledge of CS, PS, and TS semantics, as embedded in GA, enables these mappings to be precise: differing semantics are mapped to each other in such a way that loss of semantics is limited to the minimum. These mappings relate to the definition of the glue process implemented by the GA connector, which defines how a pair of producer and consumer roles coordinates in the environment. The GA glue reconciles consumer and producer roles that may differ with respect to time and space coupling as well as scoping. Hence, GA connectors support interactions among highly heterogeneous services of the Future Internet, and especially across domains.

eXtensible Service Bus: We apply the above connector abstractions to introduce an enhanced bus paradigm, the eXtensible Service Bus (XSB). XSB features richer interaction semantics than common ESB implementations to deal effectively with the increased Future Internet heterogeneity. Moreover, from its very conception, XSB incorporates special consideration for the cross-integration of heterogeneous interaction paradigms. When mapping between such paradigms, special attention is paid to the preservation of interaction semantics. XSB has the following features:

• XSB is an abstract bus that prescribes only the high-level semantics of the common bus protocol. The XSB common bus protocol features GA semantics.

• Heterogeneous systems can be plugged into the XSB by employing binding components that adapt between the native middleware of the deployed system and the common bus protocol. This adaptation is based on the systematic abstractions and mappings discussed above

• XSB, being an abstract bus, can have different implementations. This means that it needs to be complemented with a substrate which at least supports: (1) deployment (i.e., plugging) of various systems on the bus, and (2) a common bus protocol implementing GA semantics. With respect to the latter, we envision that a GA protocol realization may either be designed and built from scratch (still supposing at least an IP-based transport substrate) or be implemented by conveying GA semantics on top of an existing higher-level protocol used as transport carrier. The latter solution can be attractive, as it facilitates GA protocol realizations in different contexts and domains.

We have carried out an early realization of XSB on the PEtALS ESB. In particular, we addressed the workflow-based orchestration of heterogeneous systems, which is a preliminary step before dealing with peer-wise system integration. This work already provides a successful feasibility study of the XSB concept. This work comprises: (i) extending the BPEL workflow language with GA API primitives; (ii) introducing transformation between the GA-extended BPEL and the standard BPEL, which consists in encapsulating GA primitives into standard BPEL primitives and enables conveying GA semantics on top of BPEL primitives and subsequently on top of the common bus protocol primitives; (iii) providing templates for systematic and highly facilitated building of binding components; and (iv) introducing transformations between native system interface descriptions and GA-based interface descriptions.