Grid'5000 user report for
User information( user)
More user information in the user management interface.
- Fault-tolerance: Checkpointing overhead (Programming) [planned]
Description: The objective of this experiment is to evaluated the overhead of checkpointing of large-scale parallel application running on multiple sites.
- System for Distributed Storage (Middleware) [planned]
Description: We will perform a large-scale experiment to evaluate a system for distributed storage of data in non-dedicated resource.
- A parallel method for the thickness and optical constants estimation of thin films (Application) [in progress]
Description: We are working in a parallel method for solving the problem of estimating the thickness and optical constants of thin films. This problem is extremelly hard to solve and needs much computational processing time.
This test help us collecting data to adjust the method robustness, perform better estimations, and further reduce the computational time needed by each execution.
Each test in a sequential computer with one 3 GHz CPU takes almost 2 months to finish. In other hand, this application consumes very little network bandwidth. In a near future our objective is to evalute the behaviour of our method in a heterogeneous CPU environment. These reasons make the Grid5000 environment a nice place to do this experiments.
In each test we ran the method for aproximatelly 12 hours in 60-150 machines and we collected numerical data to help us evaluating the method.
Results: Until now we discovered and solved many problems regarding the initial aproximations used by the method, however it still needs some improvements.
- IG-Deployer: Automatic Deployment of Opportunistic Grids, using Taktuk (Other) [in progress]
Description: In order to provide InteGrade, an opportunistic grid, as a full service, a reliable, fast and scalable automatic deployment strategy is needed. Ig-Deployer is a tool for deploying integrade, based on taktuk applications. It fulfills all the necessary tasks to affiliate a new node into the an Integrade cluster. The experiments are going to be held in order to show that ig-deployer fullfils it's tasks, that it is reliable, and that the deployment time scales logarithmic in the number of nodes for a large set of notes.
Results: Yet to be tested
More information here