Michel Bagein

Objective evaluation of grapheme to phoneme conversion for text-to-speech synthesis in French

This paper reports on a cooperative international evaluation of grapheme-to-phoneme (GP) conversion for text-to-speech synthesis in French. Test methodology and test corpora are described. The results for eight systems are provided and analysed in some detail. The contribution of this paper is twofold: on the one hand, it gives an accurate picture of the state-of-the-art in the domain of GP conversion for French, and points out the problems still to be solved. On the other hand, much room is devoted to a discussion of methodological issues for this task. We hope this could help future evaluations of similar systems in other languages.

Boosting Open-Source Database Engines with Graphics Processors

In recent years, graphics chips known as GPUs have become increasingly important in the field of supercomputing. These chips are effective on heavy and highly parallelizable calculations. They naturally found applications in the scientific world for example in weather forecasting, molecular modeling and audio or video processing. Previous work has shown the ability to implement on a GPU some of the operations that can typically be found in database engines. in this context, we studied how to adapt an open source relational database engine to allow it to run on a GPU. for this, we developed a prototype based on the SQLite open source project allowing us to perform a variety of insertion and selection queries that manipulate data stored in the GPU memory. Our database engine support record filtering, relational joins, aggregation functions and most of the operators and functions working on both numbers and strings. for all of those, we implemented efficient algorithms taking into account the specific architecture of the GPU. the execution speedups of most queries were improved by a factor from ten to two hundred in comparison of the CPU counterpart. Therefore, our GPU-running database engine is able to improve system responsiveness and/or reduce the number of machines needed for an equivalent amount of data to process, thereby improving the energy efficiency.

Improving Performances of an Embedded Relational Database Management System with a Hybrid CPU/GPU Processing Engine

End-user systems are increasingly impacted by the exponential growth of data volumes and their processing. Moreover, post-processing operations, essentially dedicated to ergonomic features, require more and more resources. Improving overall performances of embedded relational database management systems (RDBMS) can contribute to deliver better responsiveness of end-user systems while increasing the energy efficiency. In this paper, it is proposed to upgrade SQLite, the most-spreaded embedded RDBMS, with a hybrid CPU/GPU processing engine combined with appropriate data management. With the proposed solution, named CuDB, massively parallel processing is combined with strategic data placement, closer to computing units. Experimental results revealed, in all cases, better performances and power efficiency compared to SQLite with an in-memory database.

Efficiency of GPUs for Relational Database Engine Processing

Relational database management systems (RDBMS) are still widely required by numerous business applications. Boosting performances without compromising functionalities represents a big challenge. To achieve this goal, we propose to boost an existing RDBMS by making it able to use hardware architectures with high memory bandwidth like GPUs. In this paper we present a solution named CuDB. We compare the performances and energy efficiency of our approach with different GPU ranges. We focus on technical specificities of GPUs which are most relevant for designing high energy efficient solutions for database processing.

Optimizing Xen Inter-domain Communications

issue for the efficiency of High Performance Computing. Virtualization adds layers and penalizes data transfers. Therefore, two new tools are proposed here to speed up data transfers between the Xen privileged virtual machine and unprivileged virtual machines. The first tool, named GNTRING, allows faster data transfers between domains by avoiding data copies, thanks to a direct access ring buffer. It offers a peak bandwidth of 500 GB/s, 390 times higher than existing solutions. The second tool, named GNTADDR, drastically reduces transferred data volume through page sharing. It is efficient for multiple transferred memory zones. Once pages are shared, data transfers come down to a signal exchange taking between 30 and 60 us.