Section: New Results

Data and Process Sharing

Hybrid P2P/cloud Architecture

Participants : Esther Pacitti, Patrick Valduriez.

Zenith adopts a hybrid P2P/cloud architecture. P2P naturally supports the collaborative nature of scientific applications, with autonomy and decentralized control. Peers can be the participants or organizations involved in collaboration and may share data and applications while keeping full control over some of their data (a major requirement for our application partners). But for very-large scale data analysis or very large workflow activities, cloud computing is appropriate as it can provide virtually infinite computing, storage and networking resources. Such hybrid architecture also enables the clean integration of the users' own computational resources with different clouds.

In [24] , we define Zenith's architecture with P2P data services and cloud data services. We model an online scientific community as a set of peers and relationships between them. The peers have their own data sources. The relationships are between any two or more peers and indicate how the peers and their data sources are related, e.g. friendship, same semantic domain, similar schema. The P2P data services include basic services (metadata and uncertain data management): recommendation, data analysis and workflow management through the Shared-data Overlay Network (SON) middleware. The cloud P2P services include data mining, content-based information retrieval and workflow execution. These services can be accessed through web services, and each peer can use the services of multiple clouds.

Social-based P2P Data Sharing

Participants : Reza Akbarinia, Emmanuel Castanier, Esther Pacitti, Didier Parigot, Patrick Valduriez, Guillaume Verger.

As a validation of the ANR DataRing project, we have developed P2PShare, a P2P system for large-scale probabilistic data sharing in scientific communities. P2PShare leverages content-based and expert-based recommendation. It is designed to manage probabilistic and deterministic data in P2P environments. It provides a flexible environment for integration of heterogeneous sources, and takes into account the social based aspects to discover high quality results for queries by privileging the data of friends (or friends of friends), who are expert on the topics related to the query.

Using the Shared-Data Overlay Network (SON), we have implemented a prototype of P2PShare that integrates three major DataRing services: ProbDB, a probabilistic database management service for relational data; WebSmatch, an environment for Web data integration; and P2Prec, a social-based P2P recommendation service for large-scale content sharing.

In [50] , , we describe the demo of P2PShare's main services, e.g., gossiping topics of interest among friends, key- word querying for contents, and probabilistic queries over datasets.

View Selection in Distributed Data Warehousing

Participants : Zohra Bellahsène, Imen Mami.

Scientific data generate large amounts of data which have to be collected and stored for analytical purpose. One way to help managing and analyzing large amounts of data is data warehousing, whereby views over data are materialized [23] . At large scale, a data warehouse can be distributed. We have examined the problem of choosing a set of views and a set of data warehouse nodes at which these views should be materialized so that the full query workload is answered with the lowest cost. To address this problem, we extended our view selection method that we proposed for the centralized case. Thus, we modelled the distributed view selection problem as a Constraint Satisfaction Problem (CSP). Furthermore, we introduced the distributed AND-OR view graph, which can be seen as an extensive form of the AND-OR view graph to reflect the relation between views and communication network within the distributed scenario. The experiment results show that our approach provides better performance compared with the genetic algorithm in term of the solution quality (i.e., the quality of the obtained set of materialized views). We demonstrated experimentally that our approach provides better results in term of cost savings when the view selection is decided under space and maintenance cost constraints [44] .

Scientific Workflow Management

Participants : Ayoub Ait Lahcen, Jonas Dias, Didier Parigot, Patrick Valduriez.

Scientific experiments based on computer simulations can be defined, executed and monitored using Scientific Workflow Management Systems (SWfMS). Several SWfMS are available, each with a different goal and a different engine. Due to the exploratory analysis, scientists need to run parameter sweep (PS) workflows, which are workflows that are invoked repeatedly using different input data. These workflows generate a large amount of tasks that are submitted to High Performance Computing (HPC) environments. Different execution models for a workflow may have significant differences in performance in HPC. However, selecting the best execution model for a given workflow is difficult due to the existence of many characteristics of the workflow that may affect the parallel execution.

In [36] , we develop a study to show performance impacts of using different execution models in running PS workflows in HPC. Our study contributes by presenting a characterization of PS workflow patterns (the basis for many existing scientific workflows) and its behavior under different execution models in HPC. We evaluated four execution models to run workflows in parallel. Our study measures the performance behavior of small, large and complex workflows among the evaluated execution models. The results can be used as a guideline to select the best model for a given scientific workflow execution in HPC. Our evaluation may also serve as a basis for workflow designers to analyze the expected behavior of an HPC workflow engine based on the characteristics of PS workflows.

This work was done in the context of the the CNPq-Inria project DatLuge and FAPERJ-Inria P2Pcloud project .

In the context of SON, we also proposed a declarative workflow language based on service/activity rules. In [27] , [46] , we present a formal approach that combines component-based development with well-understood methods and techniques from the field of Attribute Grammars and Data-Flow Analysis in order to specify the behavior of P2P applications, and then construct an abstract representation (i.e., Data-Dependency Graph) to perform analyzes on it. This formal approach makes it possible to infer a dependency graph for SON applications that provides for automatic parallelization.

Plants identification and classification from social image data

Participants : Hervé Goëau, Alexis Joly, Saloua Litayem.

This work is done in collaboration with the botanists of the AMAP UMR team (CIRAD) and with Inria team IMEDIA. Inspired by citizen sciences, the main goal of this trans-disciplinary work is to speed up the collection and integration of raw botanical observation data, while providing to potential users an easy and efficient access to this botanical knowledge. We therefore did continue working intensively on plants identification and classification [54] , [37] , [38] , [26] . We first developed a new interactive method [37] for the visual identification of plants from social image data. Contrary to previous content-based identification methods and systems that mainly relied on leaves, or in few other cases on flowers, it makes use of five different organs and plant's views including habit, flowers, fruits, leaves and bark. Thanks to an interactive query widget, the tagging process of the different organs and views is as simple as drag-and-drop operations and does not require any expertise in botany. All training pictures used by the system were continuously collected during one year through a crowdsourcing application and more than 17K images are now integrated. System-oriented and human-centered evaluations of the application show that the results are already satisfactory and therefore very promising in the long term to identify a richer flora.

Besides, we did continue working on leaf-based identification notably through the organization of and participation to ImageCLEF plant identification evaluation campaign 2012 [54] .

Finally we did apply one of our former work related to multi-source shared-nearest neighbors clustering to an original experiment aimed at evaluating if we were able to automatically recover morphological classifications built by the botanists themselves [38] . The results are very promising, since all clusters discovered automatically could be easily matched to one node of a morphological tree built by botanists.