Grid middleware systems provide a large set of services to deal with resources like information services, resources reservation/allocation, file transfert and job launching and monitoring. They are independent of any programming model.
Deploying manually an application means interacting with all these services: this is very complex and requires grid experts.
In order to achieve the invisible grid, it is important to be able to automatically deploy an application. A grid user should only provide a description of his/her application as well as a quality of service it expect from the execution of the application.
ADAGE is our activity related to the study of the automatization of the deployment of applications on computation grids. The first version of ADAGE only deals with static applications while the ongoing one targets dynamic application.
The architecture of ADAGE is depicted in the Figure on the right. It contains three stages. First, the input information are collected: they are made of the description of the application, the description of the quality of the execution requests, and a description the ressources or where they can be found (like through MDS).
ADAGE is internally based on a generic application description model (GADe) so as to support any kind of programming model and/or to support multi-programming model applications. The conversion to GADe is achieved thanks to a dedicated plugin. Currently, the following programming model are supported:
Second, there is the scheduling stage whose goal is to select a set of ressources and to map the element of the application to theses resources while satisfaying the constraints like operating system, processors, memory, etc. A deployment plan is computed according to a selected scheduler. Schedulers are plugins too in ADAGE. Currently, available plugins areonly basic ones like round robin and random have been implemented:
Third, the deployment plan is executed. At this step, files (executables and data files) are transferred accordingly to the deployment plan, then, processes are created, configured and launched and finally an execution report is generated. Currently, ADAGE makes use of an API that abstracts the grid middleware technology like OAR.
ADAGE development is hosted on the INRIA forge. You may be interested first by this HOWTO install, then by this user doc. A list of implemented and planned features is available here. People interested in supporting their own programming model should read this plugin-writer HOWTO.
More information on how to communicate with the adage daemon in this page.
Co-deployment and re-deployment of grid-based applications is also possible thanks to the CoRDAGe project.
For any information and support about ADAGE, a mailing-list is available here