Section: New Results
Specific studies: Web services orchestrations
Participants : Ajay Kattepur, Albert Benveniste, Claude Jard.
Web services orchestrations and choreographies refer to the composition of several Web services to perform a co-ordinated, typically more complex task. We decided to base our study on a simple and clean formalism for WS orchestrations, namely the Orc formalism proposed by Jayadev Misra and William Cook  .
Main challenges related to Web services QoS (Quality of Service) include: 1/ To model and quantify the QoS of a service. 2/ To establish a relation between the QoS of queried Web services and that of the orchestration (contract composition); 3/ To monitor and detect the breaching of a QoS contract, possibly leading to a reconfiguration of the orchestration. Typically, the QoS of a service is modeled by a contract (or Service Level Agreement, SLA) between the provider and consumer of a given service. To account for variability. In previous years, we proposed soft probabilistic contracts specified as probabilistic distributions involving the different QoS parameters; we studied contract composition for such contracts; we developed probabilistic QoS contract monitoring; and we studied the monotonicity of orchestrations; an orchestration is monotonic if a called service improves its performance, then so does the overall orchestration.
This year, in the framework of the Associated Team FOSSA with the University of Texas at Austin (John Thywissen (PhD), Jayadev Misra and William Cook), we have extended our approach to general QoS parameters, i.e., beyond response time. We now encompass composite parameters, which are thus only partially, not totally, ordered. We have developed a general algebra to capture how QoS parameters are transformed while traversing the orchestration and we have extended our study of monotonicity. Finally, we have developed corresponding contract composition procedures. John Thywissen (from UT Austin) and Ajay Kattepur have started extending the Orc language and execution engine to support QoS according to our theory. This extension mainly consists in 1/ providing a rich type system to declare QoS domains and related algebra, and 2/ providing a new operator for Orc that allows for selecting competing returns from different sites on the basis of their QoS. A journal paper is under revision.
A key task in extending Orc for QoS was to extend the Orc engine so that causalities between the different site calls are made explicit at run time while execution progresses. This benefits from our previous work on Orc semantics, but a new set of rules has been proposed to generate causalities in an efficient way, by covering new features of the language. This is joint work of Claude Jard, Ajay Kattepur and John Thywissen from Austin. A publication is in preparation.
Besides this main line of work, other topics have been addressed by Ajay Kattepur as part of his thesis.
In  , we study variability of composite services. We model variability as a feature diagram (FD) that captures all valid configurations of its orchestration. Then, we apply pair-wise testing to sample the set of all possible configurations to obtain a concise subset. Finally, we test the composite service for selected pairwise configurations for a variety of QoS metrics such as response time, data quality, and availability. Using two case studies, Car crash crisis management and e-Health management, we demonstrate that pairwise generation effectively samples the full range of QoS variations in a dynamic orchestration. The pairwise sampling technique eliminates over 99% redundancy in configurations, while still calling all atomic services at least once.
Web services orchestrations conventionally employ exhaustive comparison of runtime quality of service (QoS) metrics for decision making. The ability to incorporate more complex mathematical packages is needed, especially in case of workflows for resource allocation and queuing systems. By modeling such optimization routines as service calls within orchestration specifications, techniques such as linear programming can be conveniently invoked by non-specialist workflow designers. Leveraging on previously developed QoS theory, we propose the use of a high-level flexible query procedure for embedding optimizations in languages such as Orc. The Optima site provides an extension to the sorting and pruning operations currently employed in Orc. Further, the lack of an objective technique for consolidating QoS metrics is a problem in identifying suitable cost functions. We use the analytical hierarchy process (AHP) to generate a total ordering of QoS metrics across various domains. With constructs for ensuring consistency over subjective judgements, the AHP provides a suitable technique for producing objective cost functions. Using the Dell Supply Chain example, we demonstrate the feasibility of decision making through optimization routines, specially when the control flow is QoS dependent. This work was published in  .
With web services quality of service (QoS) modeled as random variables, the accuracy of sampled values for precise service level agreements (SLAs) come into question. Samples with lower spread are more accurate for calculating contractual obligations, which is typically not the case for web services QoS. Moreover, the extreme values in case of heavy-tailed distributions (eg. 99.99 percentile) are seldom observed through limited sampling schemes. To improve the accuracy of contracts, we propose the use of variance reduction techniques such as importance sampling. We demonstrate this for contracts involving demand and refuel operations within the Dell supply chain example. Using measured values, efficient forecasting of future deviation of contracts may also be performed. A consequence of this is a more precise definition of sampling, measurement and variance tolerance in SLA declarations. This work was published in  .