Federation and abstraction of heterogeneous global computing platforms with the YML framework

International audienceGlobal computing platforms have become popular tools for the resolution of large scale problems. They are often independent without any interoperability between each other. Therefore, clients are now asking for a better availability and scalability. In previous work, we presented the YML framework which was a first attempt to enable the development and deployment of applications on several global computing middleware. However, it suffered from scalability issues and a static approach. In this paper, we highlight recent extensions of YML. We achieve a dynamic federation of computing middleware because YML is now able to manage at the run-time several middleware back-ends. Other improvements such as an OmniRPC back-end, a component binary cache mechanism, a new data management module, and a new data type supported by the YML front-end, yield to a much better scalability. We present the first evaluations of these extensions on a network of workstations and with a typical distributed sort application. Although the results show a significant overhead, they stress the benefits of binary caching and dynamic back-ends federation