International audience

Summary Distributed systems for big data management very often face the problem of load imbalance among nodes. To address this issue, there exist almost as many load balancing strategies as there are different systems. When designing a scalable distributed system geared towards handling large amounts of information, it is often not so easy to anticipate which kind of strategy will be the most efficient to maintain adequate performance regarding response time, scalability, and reliability at any time. Based on this observation, we describe a generic API to implement and experiment any strategy independently from the rest of the code, prior to a definitive choice for instance. We then show how existing load balancing strategies used by famous systems could be implemented with this API. We also present how this work has helped us implement load balancing on our distributed system and modify the behavior of our strategy in a few lines of code. This led us to easily perform various experiments to determine the most efficient scheme for our system. This paper is an extension to our work presented at Workshop on Parallel and Distributed Computing for Big Data Applications (WPBA) 2014. We detail here more experiments and extend the use of the API to a broad class of big data storage systems. Copyright © 2015 John Wiley & Sons, Ltd.

Component-oriented software has become a useful tool to build larger and more complex systems by describing the application in terms of encapsulated, loosely coupled entities called components. At the same time, asynchronous programming patterns allow for the development of efficient distributed applications. While several component models and frameworks have been proposed, most of them tightly integrate the component model with the middleware they run upon. This intertwining is generally implicit and not discussed, leading to entangled, hard to maintain code. This article describes our efforts in the development of the GCM/ProActive framework for providing distributed and adaptable autonomous components. GCM/ProActive integrates a component model designed for execution on large-scale environments, with a programming model based on active objects allowing a high degree of distribution and concurrency. This new integrated model provides a more powerful development, composition, and execution environment than other distributed component frameworks. We illustrate that GCM/ProActive is particularly adapted to the programming of autonomic component systems, and to the integration into a service-oriented environment. Copyright © 2014 John Wiley & Sons, Ltd.

This article presents a GPU adaptation of a specific Monte Carlo and classification based method for pricing American basket options, due to Picazo. Some optimizations are exposed to get good performance of our parallel algorithm on GPU. In order to benefit from different GPU devices, a dynamic strategy of kernel calibration is proposed. Future work is geared towards the use of distributed computing infrastructures such as Grids and Clouds, equipped with GPUs, in order to benefit for even more parallelism in solving such computing intensive problem in mathematical finance.

Improving performance of Web services interactions is an important factor to burst the adoption of SOAin mission-critical applications, especially when they deal with large business objects whose transfer time is not negligible. Designing messages dynamic granularity (offloading) is a key challenge for achieving good performances. This requires the server being able to predict the pieces of data actually used by clients in order to send only such data. However, exact prediction is not easy, and consequently lazy interactions are needed to transfer additional data whenever the prediction fails. To preserve semantics, lazy accesses to the results of a Web service interaction need to work on a dedicated copy of the business object stored as application state. Thus, dynamic offloading can experience an overhead due to a prediction failure, which is the sum of round-trip and storage access delays, which could compromise the benefits of the technique. This paper improves our previous work enabling dynamic offloading for both IN and OUT parameters, and analyses how attributes copies impact on the technique, by comparing the overheads introduced by different storage technologies in a real implementation of a Web services framework that extends CXF. More specifically, we quantitatively characterize the execution contexts that make dynamic offloading effective, and the expected accuracy of the predictive strategy to have a gain in term of response time compared to plain services invocations. Finally, the paper introduces the Attribute Loading Delegation technique that enables optimized data-transfers for those applications where data-intensive multiple-interactions take place.

SUMMARY Current solutions for hybrid grid/cloud computing have been developed to hide from heterogeneity, dynamicity, and distributed nature of resources. These solutions are, however, insufficient to support distributed applications with non‐trivial communication patterns among processes or that are structured so as to reflect the organization of resources they are deployed onto. In this paper, we present a generic, adaptable, and extensible component‐based middleware that seamlessly enables a transition of non‐trivial applications from traditional grids to hybrid grid–cloud platforms. This middleware goes beyond the resolution of well‐known technical challenges for multi‐domain computing, as it offers mechanisms to exploit the hierarchical, heterogeneous, and dynamic natures of platforms. We validate its capabilities and versatility through two use cases: an Internet‐wide federation of Distributed Service Buses and a runtime supporting DD high‐performance computing in heterogeneous computing environments using programming that is similar to message‐passing interface. Performance results show the efficiency and usefulness of our middleware and so contribute to promote research efforts geared toward flexible, on‐demand information technology solutions. Copyright © 2012 John Wiley & Sons, Ltd.

Business Process Outsourcing is nowadays common in companies. In particular, the distribution of such business processes encompasses the inclusion of external service providers in the overall process as well as the usage of external infrastructures like clouds. Both of these approaches lead to decentralization and outsourcing of a part of the global workflow, resulting in a complexified management of the global orchestration. As a matter of fact, the overall management data are decentralized among different domains and must, most of the time, be gathered manually. This paper presents a framework that eases multi-domain orchestration management. Our approach extracts, gathers and digests data from the decentralized processes in order to provide an unified and global view of a distributed orchestration.

Distributed computing environments have evolved from in-house clusters to Grids and now Cloud platforms. We, as others, provide HPC benchmarks results over Amazon EC2 that show a lower performance of Cloud resources compared to private resources., So, it is not yet clear how much of impact Clouds will have in high performance computing (HPC). But hybrid Grid/Cloud computing may offer opportunities to increase overall applications performance, while benefiting from in-house computational resources extending them by Cloud ones only whenever needed. In this paper, we advocate the usage of Proactive, a well established middleware in the grid community, for mixed Grid/Cloud computing, extended with features to address Grid/Cloud issues with little or, no effort for application developers. We also introduce a framework, developed in the context of the Disco Grid project, based upon the Proactive middleware to couple HPC domain-decomposition SPMD applications in heterogeneous multi-domain environments. Performance results, coupling Grid and Cloud resources for the execution of such, kind of highly communicating and processing intensive applications, have shown an overhead of about 15%, which is a non-negligible value, but lower enough to consider using such environments to achieve a better cost-performance trade-off than using exclusively Cloud resources.

We propose to address collective communications in distributed components through collective interfaces. Collective interfaces handle data distribution, parallelism and synchronization, and they expose collective behaviors in the definition of components. We show, as an illustration, that collective interfaces allow the encoding of SPMD programming in a better structured and less error prone way. We verify the scalability and performance of collective interfaces in an experiment on up to 100 machines.

While many dream and talk about service level agreement (SM) and quality of service (QoS) for service oriented architectures (SOA), the practical reality of grid computing are still far from providing effective techniques enabling such contractual agreements. Towards this goal, this paper provides an overview of the techniques offered by ProActive to set and use contractual agreements. Based on the identification of roles, application developer, infrastructure manager, application user, the actors of a grid environment can specify what is required or what is provided at various levels. The results are both flexibility and adaptability, matching the application constraints and the environment characteristics with various techniques.

Component-based development has emerged as an effective approach to building flexible systems, but there is little experience in applying this approach to grid programming. This paper presents our experience with reengineering a high performance numerical solver to become a component-based grid application. The adopted component model is an extension of the generic fractal model that specifically targets grid environments. The paper provides qualitative and quantitative evidence that componentisation has improved the modifiability and reusability of the application while not significantly affecting performance.

No abstract available.

In this paper, we present a proposal for hierarchical MPI programming through some intuitive extensions to the MPI standard that may help users to develop non-embarrassingly parallel grid applications in a topology- aware manner. Afterwards, we present the design of such a support based upon a component model suited to grid computing (the EU CoreGRID grid component model - GCM - and its implementation in the ProActive grid environment) to handle inter-cluster and group communications. The usage of such components to handle high-level data distribution, parallelism and synchronization seems to be the most adequate technology to support MPI primitives in multi-cluster grids as they provide a built-in support to the encapsulation of native code, collective interfaces, tunneling of communications and a hierarchical and adaptable structure. The preliminary results have shown that the overhead is not negligible, but within the expected range. However we can expect the benefits to applications to bypass the generated overhead.

This paper introduces a Grid software architecture offering fault tolerance, dynamic and aggressive load balancing and two complementary parallel programming paradigms. Experiments with financial applications on a real multi-site Grid assess this solution. This architecture has been designed to run industrial and financial applications, that are frequently time constrained and CPU consuming, feature both tightly and loosely coupled parallelism requiring generic programming paradigm, and adopt client-server business architecture.

This article presents an evolution of classical SPMD programming for clusters and grids. It is named "object-oriented SPMD" as it is based on remote method invocation. More precisely, it is based on an active object pattern, extended as a typed group of active objects, to which SPMD's specificities are added. The proposed programming model is more flexible: techniques to postpone barrier and to remove any explicit loop make it possible to privilege reactivity and reuse. The resulting OO-SPMD API has been implemented in ProActive. Good scalability and quite competitive performances, compared to what is obtained using C-MPI, are demonstrated.

Summary form only given. Within the trend of object-based distributed computing, we present the design and implementation of a numerical simulation for electromagnetic waves propagation. A sequential Java design and implementation is first presented. Further, a distributed and parallel version is derived from the first, using an active object pattern. In addition, benchmarks are presented on this nonembarrassingly parallel application. A first contribution resides in the sequential object-oriented design that proved to be very modular and extensible; the classes and abstractions are designed to allow both element and volume type methods, furthermore, valid on structured, unstructured, or hybrid meshes. Compared to a Fortran version, the performance of this highly modular version proved to be in the same range. It is also shown how smoothly the sequential version can be distributed, keeping the same structuring and object abstractions, allowing to deal with larger data size. Finally, benchmarks on up to 64 processors compare the performances with respect to sequential and parallel versions, putting that in perspective with a comparable Fortran version.

Increasing complexity of distributed applications and commodity of resources through grids are making the tasks of deploying those applications harder. There is a clear need for standard tools allowing versatile deployment and analysis of distributed applications. We present here a solution for the deployment and monitoring of applications written using ProActive, an experimental Java-based library for concurrent, distributed and mobile computing. We describe the use of XML-based descriptor for the deployment part of a distributed application and the use of IC2D (Interactive Control and Debugging of Distribution), for the monitoring and steering of the running application. Those ideas, concepts, and experiments are a contribution towards the construction of integrated environments for component-based grid programming.